PageSourceSearch

https://thedebrief.ai/archive

html thedebrief.ai collected 2026-10-05 15:09:08 UTC 538,613 bytes, 1 lines download raw bytes

1<!DOCTYPE html><html lang="en"><head><meta charSet="utf-8"/><meta name="viewport" content="width=device-width, initial-scale=1"/><link rel="preload" as="image" href="/logo.png"/><link rel="stylesheet" href="/assets/styles-jpKyklsl.css" data-precedence="default"/>
1<script async="" src="https://js.sparkloop.app/embed.js?publication_id=pub_8c86ac7f847e" data-sparkloop="true"></script>
1<title>Archive - The AI Debrief</title><meta name="author" content="Drew Collins"/><meta name="google-site-verification" content="MU_bV6Qipbe9-23mWqinkrEKZeUlRun_WjerFbHBXJs"/><meta name="twitter:card" content="summary_large_image"/><meta name="description" content="Every issue of The AI Debrief - a twice-weekly AI newsletter for operators. New issues land here twice a week."/><meta property="og:site_name" content="The AI Debrief"/><meta property="og:title" content="Archive - The AI Debrief"/><meta property="og:description" content="Every issue of The AI Debrief - a twice-weekly AI newsletter for operators. New issues land here twice a week."/><meta property="og:type" content="website"/><meta property="og:url" content="https://thedebrief.ai/archive"/><meta property="og:image" content="https://thedebrief.ai/og.jpg"/><meta name="twitter:title" content="Archive - The AI Debrief"/><meta name="twitter:description" content="Every issue of The AI Debrief - a twice-weekly AI newsletter for operators. New issues land here twice a week."/><meta name="twitter:image" content="https://thedebrief.ai/og.jpg"/><link rel="modulepreload" href="/assets/index-Dx_NznIf.js"/><link rel="modulepreload" href="/assets/email-suggestion-BEKGUyL0.js"/><link rel="modulepreload" href="/assets/preload-helper-CvAz7yeD.js"/><link rel="modulepreload" href="/assets/useSearch-BLga_yIz.js"/><link rel="modulepreload" href="/assets/archive.index-B7lBVB9C.js"/><link rel="modulepreload" href="/assets/layout-CS1Tl4bZ.js"/><link rel="modulepreload" href="/assets/posts-BifpePUO.js"/><link rel="icon" href="/favicon.png" type="image/png"/><link rel="apple-touch-icon" href="/logo.png"/><link rel="canonical" href="https://thedebrief.ai/archive"/>
1<script type="application/ld+json">{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://thedebrief.ai/#organization","name":"The AI Debrief","url":"https://thedebrief.ai/","logo":"https://thedebrief.ai/logo.png","description":"An independent twice-weekly AI newsletter for operators, covering practical AI workflows, tools and automation.","founder":{"@id":"https://thedebrief.ai/#drew"}},{"@type":"Person","@id":"https://thedebrief.ai/#drew","name":"Drew Collins","jobTitle":"Writer, The AI Debrief","url":"https://thedebrief.ai/"}]}</script>
1<script type="application/ld+json">{"@context":"https://schema.org","@type":"CollectionPage","name":"Archive - The AI Debrief","description":"Every issue of The AI Debrief - a twice-weekly AI newsletter for operators. New issues land here twice a week.","url":"https://thedebrief.ai/archive","isPartOf":{"@type":"WebSite","name":"The AI Debrief","url":"https://thedebrief.ai/"},"mainEntity":{"@type":"ItemList","itemListElement":[{"@type":"ListItem","position":1,"url":"https://thedebrief.ai/p/gstack-can-now-run-your-eng-team","name":"Gstack Can Now Run Your Eng Team"},{"@type":"ListItem","position":2,"url":"https://thedebrief.ai/p/gpt-6-sol-skips-chat-work-only","name":"GPT-6 Sol Skips Chat. Work Only."},{"@type":"ListItem","position":3,"url":"https://thedebrief.ai/p/the-10-cent-model-has-a-100-second-catch","name":"The 10-cent model has a 100-second catch"},{"@type":"ListItem","position":4,"url":"https://thedebrief.ai/p/your-coding-agent-pinned-a-hash-it-got-a-branch","name":"Your coding agent pinned a hash. It got a branch."},{"@type":"ListItem","position":5,"url":"https://thedebrief.ai/p/google-s-1-38-voice-agent-has-a-catch","name":"Google's $1.38 voice agent has a catch"},{"@type":"ListItem","position":6,"url":"https://thedebrief.ai/p/openai-wants-to-run-your-agents","name":"OpenAI wants to run your agents"},{"@type":"ListItem","position":7,"url":"https://thedebrief.ai/p/the-agents-got-out-of-the-test","name":"The agents got out of the test"},{"@type":"ListItem","position":8,"url":"https://thedebrief.ai/p/gemini-stopped-watching-the-whole-video","name":"Gemini stopped watching the whole video"},{"@type":"ListItem","position":9,"url":"https://thedebrief.ai/p/cursor-just-lost-openai","name":"Cursor just lost OpenAI"},{"@type":"ListItem","position":10,"url":"https://thedebrief.ai/p/the-16-tool-you-can-delete","name":"The $16 tool you can delete"},{"@type":"ListItem","position":11,"url":"https://thedebrief.ai/p/three-of-these-are-safe-to-ignore-25561566","name":"Three of these are safe to ignore"},{"@type":"ListItem","position":12,"url":"https://thedebrief.ai/p/three-of-these-are-safe-to-ignore","name":"Three of these are safe to ignore"},{"@type":"ListItem","position":13,"url":"https://thedebrief.ai/p/30-days-10-minutes-a-day-25496488","name":"30 days. 10 minutes a day."},{"@type":"ListItem","position":14,"url":"https://thedebrief.ai/p/30-days-10-minutes-a-day","name":"30 days. 10 minutes a day."},{"@type":"ListItem","position":15,"url":"https://thedebrief.ai/p/google-halved-the-price-until-january-25495572","name":"Google halved the price. Until January."},{"@type":"ListItem","position":16,"url":"https://thedebrief.ai/p/google-halved-the-price-until-january","name":"Google halved the price. Until January."},{"@type":"ListItem","position":17,"url":"https://thedebrief.ai/p/the-100-agent-is-now-20","name":"The $100 agent is now $20"},{"@type":"ListItem","position":18,"url":"https://thedebrief.ai/p/the-agent-model-that-fits-on-one-gpu","name":"The agent model that fits on one GPU"},{"@type":"ListItem","position":19,"url":"https://thedebrief.ai/p/anthropic-just-filed-to-go-public-openai-is-next","name":"Anthropic just filed to go public (OpenAI is next)"},{"@type":"ListItem","position":20,"url":"https://thedebrief.ai/p/the-pope-just-picked-a-side-in-the-ai-war","name":"The Pope just picked a side in the AI war"},{"@type":"ListItem","position":21,"url":"https://thedebrief.ai/p/an-openai-founder-just-defected-to-anthropic","name":"An OpenAI founder just defected to Anthropic"},{"@type":"ListItem","position":22,"url":"https://thedebrief.ai/p/the-fda-for-ai","name":"The FDA for AI"},{"@type":"ListItem","position":23,"url":"https://thedebrief.ai/p/musk-wants-130-billion-back","name":"Musk Wants $130 Billion Back"},{"@type":"ListItem","position":24,"url":"https://thedebrief.ai/p/the-ai-arms-race-just-got-personal","name":"The AI Arms Race Just Got Personal"},{"@type":"ListItem","position":25,"url":"https://thedebrief.ai/p/florida-just-came-for-openai","name":"Florida Just Came For OpenAI"},{"@type":"ListItem","position":26,"url":"https://thedebrief.ai/p/anthropic-just-taught-ai-to-hack-the-world","name":"Anthropic Just Taught AI to Hack the World"},{"@type":"ListItem","position":27,"url":"https://thedebrief.ai/p/the-week-ai-s-security-perimeter-collapsed","name":"The Week AI's Security Perimeter Collapsed"},{"@type":"ListItem","position":28,"url":"https://thedebrief.ai/p/the-lobster-just-changed-everything","name":"The Lobster Just Changed Everything"},{"@type":"ListItem","position":29,"url":"https://thedebrief.ai/p/1-trillion-90-days-and-the-ai-trial-of-the-century","name":"$1 Trillion. 90 Days. And the AI Trial of the Century."},{"@type":"ListItem","position":30,"url":"https://thedebrief.ai/p/anthropic-sues-the-pentagon","name":"Anthropic Sues the Pentagon"},{"@type":"ListItem","position":31,"url":"https://thedebrief.ai/p/pentagon-ultimatum-shakes-ai-industry","name":"Pentagon Ultimatum Shakes AI Industry"},{"@type":"ListItem","position":32,"url":"https://thedebrief.ai/p/ai-takes-over-super-bowl-lx","name":"AI Takes Over Super Bowl LX..."},{"@type":"ListItem","position":33,"url":"https://thedebrief.ai/p/the-lobster-uprising","name":"The Lobster Uprising"},{"@type":"ListItem","position":34,"url":"https://thedebrief.ai/p/the-saaspocal
1ypse-arrives","name":"The \"SaaSpocalypse\" Arrives"}]}}</script>
1<script defer src="/~flock.js" data-proxy-url="/~api/analytics"></script>
1</head><body><!--$--><div class="flex min-h-screen flex-col bg-background"><header class="fixed inset-x-0 top-0 z-50 border-b border-white/10 bg-[#0A0A0A]"><div class="mx-auto w-full max-w-[1200px] px-5 sm:px-8 flex items-center justify-between gap-4 py-4"><a class="flex shrink-0 items-center gap-2.5 text-[oklch(0.985_0.005_270)]" href="/"><img src="/logo.png" alt="The AI Debrief" width="28" height="28" class="shrink-0 rounded-[6px]" style="width:28px;height:28px"/><span class="wordmark text-sm font-bold tracking-tight sm:text-base">The AI Debrief</span></a><div class="hidden items-center gap-6 md:flex"><nav class="flex items-center gap-5 text-xs font-bold tracking-tight text-[oklch(0.985_0.005_270)]/70 lg:gap-7"><a href="https://www.skool.com/try-ai-academy/about" target="_blank" rel="noopener noreferrer" class="transition-colors hover:text-[oklch(0.985_0.005_270)]">The AI Academy</a><a class="transition-colors hover:text-[oklch(0.985_0.005_270)]" href="/work-with-us">Work With Us</a><a class="transition-colors hover:text-[oklch(0.985_0.005_270)] active" href="/archive" data-status="active" aria-current="page">Archive</a><a class="transition-colors hover:text-[oklch(0.985_0.005_270)]" href="/resources">Resources</a><a class="transition-colors hover:text-[oklch(0.985_0.005_270)]" href="/perks">Perks</a></nav><div class="flex items-center gap-2.5"><a class="group relative overflow-hidden btn-primary btn-sm " href="/join"><span class="relative z-10 transition-transform duration-200 ease-out group-hover:-translate-x-0.5">Subscribe</span><span class="relative z-10 ml-1.5 inline-flex h-0 w-0 items-center justify-center overflow-hidden opacity-0 transition-all duration-200 ease-out group-hover:ml-1.5 group-hover:h-3.5 group-hover:w-3.5 group-hover:opacity-100"><svg xmlns="http://www.w3.org/2000/svg" width="24" height="24" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-arrow-up-right h-3.5 w-3.5" aria-hidden="true"><path d="M7 7h10v10"></path><path d="M7 17 17 7"></path></svg></span></a></div></div><button type="button" aria-label="Open menu" aria-expanded="false" class="-mr-2 flex h-11 w-11 flex-col items-center justify-center gap-1.5 md:hidden"><span class="block h-px w-6 bg-[oklch(0.985_0.005_270)]"></span><span class="block h-px w-6 bg-[oklch(0.985_0.005_270)]"></span></button></div></header><main class="flex-1"><section class="bg-background"><div class="mx-auto w-full max-w-[1200px] px-5 sm:px-8 pb-14 pt-28 text-center sm:pb-16 sm:pt-36"><h1 class="rise text-3xl font-bold tracking-tight text-foreground sm:text-4xl">Archive</h1><p class="rise mx-auto mt-4 max-w-[62ch] text-base leading-relaxed text-foreground/65 sm:text-lg" style="--reveal-delay:90ms">Every edition of The AI Debrief. Practical workflows, tools, and playbooks for founders and business leaders.</p></div></section><section class="section-light relative overflow-hidden"><div class="mx-auto w-full max-w-[1200px] px-5 sm:px-8 relative py-10 pb-20 sm:py-16 sm:pb-28"><div class="relative"><div class="mb-6 flex flex-wrap items-center justify-between gap-x-4 gap-y-3 border-b border-black/10 pb-3"><div class="flex flex-wrap items-center gap-1 sm:gap-2"><button type="button" aria-pressed="true" class="inline-flex min-h-[44px] items-center rounded-[4px] px-3 font-mono text-[11px] font-bold uppercase tracking-[0.14em] transition-colors" style="color:#141414;background-color:color-mix(in srgb, #141414 10%, transparent);box-shadow:inset 0 -2px 0 #141414">All<!-- --> (<!-- -->34<!-- -->)</button><button type="button" aria-pressed="false" class="inline-flex min-h-[44px] items-center rounded-[4px] px-3 font-mono text-[11px] font-bold uppercase tracking-[0.14em] transition-colors" style="color:rgba(20,20,20,0.5)">The Debrief<!-- --> (<!-- -->15<!-- -->)</button><button type="button" aria-pressed="false" class="inline-flex min-h-[44px] items-center rounded-[4px] px-3 font-mono text-[11px] font-bold uppercase tracking-[0.14em] transition-colors" style="color:rgba(20,20,20,0.5)">The Toolkit<!-- --> (<!-- -->19<!-- -->)</button></div><p class="font-mono text-[10px] uppercase tracking-[0.2em] text-[#141414]/45">34<!-- --> <!-- -->issues</p></div><div class="rack grid grid-cols-1 items-stretch gap-x-7 gap-y-8 sm:grid-cols-2 lg:grid-cols-3"><a href="/p/gstack-can-now-run-your-eng-team" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-toolkit)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-toolkit);background-color:color-mix(in srgb, var(--edition-toolkit) 10%, transparent)">The Toolkit</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">October 2, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">Gstack Can Now Run Your Eng Team</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">Garry Tan&#x27;s opinionated eng-team pack for Claude Code. Free, MIT, and easy to miss until you add one section to CLAUDE.md. Here is the install, the trap, and the 10-minute start.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-toolkit)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/gpt-6-sol-skips-chat-work-only" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-debrief)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-debrief);background-color:color-mix(in srgb, var(--edition-debrief) 10%, transparent)">The Debrief</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">September 29, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">GPT-6 Sol Skips Chat. Work Only.</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">Two defaults shipped in one day. Sol is in Work, Codex, and the API, not Chat. Opus is on every Anthropic surface with harder gates. Pick Monday&#x27;
1s default from the placement, not the press release.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-debrief)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/the-10-cent-model-has-a-100-second-catch" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-toolkit)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-toolkit);background-color:color-mix(in srgb, var(--edition-toolkit) 10%, transparent)">The Toolkit</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">September 25, 2026</span></div><span class="mt-4 block font-mono text-[10px] font-bold uppercase tracking-[0.2em]" style="color:var(--edition-toolkit)">No. <!-- -->29</span><h3 class="mt-2 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">The 10-cent model has a 100-second catch</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">Independently measured, it is the cheapest model anywhere near its score. At the setting that earns that score it thinks for 100 seconds before it says a word. Here is what to move to it, and the 2...</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-toolkit)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/your-coding-agent-pinned-a-hash-it-got-a-branch" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-debrief)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-debrief);background-color:color-mix(in srgb, var(--edition-debrief) 10%, transparent)">The Debrief</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">September 22, 2026</span></div><span class="mt-4 block font-mono text-[10px] font-bold uppercase tracking-[0.2em]" style="color:var(--edition-debrief)">No. <!-- -->28</span><h3 class="mt-2 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">Your coding agent pinned a hash. It got a branch.</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">A zero-click hole in every major coding agent&#x27;s plugin loader. Two vendors fixed it months ago, two have not, and the fix is a version number.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-debrief)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/google-s-1-38-voice-agent-has-a-catch" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-toolkit)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-toolkit);background-color:color-mix(in srgb, var(--edition-toolkit) 10%, transparent)">The Toolkit</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">September 18, 2026</span></div><span class="mt-4 block font-mono text-[10px] font-bold uppercase tracking-[0.2em]" style="color:var(--edition-toolkit)">No. <!-- -->27</span>
1<h3 class="mt-2 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">Google&#x27;s $1.38 voice agent has a catch</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">Gemini 3.8 Live is free to try and less than half of OpenAI&#x27;s price. The version that tops the leaderboard is not the cheap one. Here&#x27;s what to build on it anyway.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-toolkit)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/openai-wants-to-run-your-agents" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-debrief)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-debrief);background-color:color-mix(in srgb, var(--edition-debrief) 10%, transparent)">The Debrief</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">September 15, 2026</span></div><span class="mt-4 block font-mono text-[10px] font-bold uppercase tracking-[0.2em]" style="color:var(--edition-debrief)">No. <!-- -->26</span><h3 class="mt-2 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">OpenAI wants to run your agents</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">The whole agent loop is now one API call with no platform fee. Two lines in the docs decide whether you can actually use it.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-debrief)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/the-agents-got-out-of-the-test" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-debrief)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-debrief);background-color:color-mix(in srgb, var(--edition-debrief) 10%, transparent)">The Debrief</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">September 10, 2026</span></div><span class="mt-4 block font-mono text-[10px] font-bold uppercase tracking-[0.2em]" style="color:var(--edition-debrief)">No. <!-- -->25</span><h3 class="mt-2 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">The agents got out of the test</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">Anthropic and OpenAI both admitted this week that agents under evaluation reached real systems. The sandbox is whatever permissions you handed over.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-debrief)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/gemini-stopped-watching-the-whole-video" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-toolkit)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-toolkit);background-color:color-mix(in srgb, var(--edition-toolkit) 10%, transparent)">The Toolkit</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">September 4, 2026</span></div><span class="mt-4 block font-mono text-[10px] font-bold uppercase tracking-[0.2em]" style="color:var(--edition-toolkit)">No. <!-- -->24</span>
1<h3 class="mt-2 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">Gemini stopped watching the whole video</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">Google&#x27;s new video mode reads the transcript first and only pulls the frames it needs. Here&#x27;s the notes machine to build on top of it.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-toolkit)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/cursor-just-lost-openai" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-debrief)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-debrief);background-color:color-mix(in srgb, var(--edition-debrief) 10%, transparent)">The Debrief</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">September 1, 2026</span></div><span class="mt-4 block font-mono text-[10px] font-bold uppercase tracking-[0.2em]" style="color:var(--edition-debrief)">No. <!-- -->23</span><h3 class="mt-2 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">Cursor just lost OpenAI</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">OpenAI gave Cursor 76 days&#x27; notice. The lesson applies to every tool that rents its brain from somebody else.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-debrief)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/the-16-tool-you-can-delete" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-toolkit)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-toolkit);background-color:color-mix(in srgb, var(--edition-toolkit) 10%, transparent)">The Toolkit</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">August 28, 2026</span></div><span class="mt-4 block font-mono text-[10px] font-bold uppercase tracking-[0.2em]" style="color:var(--edition-toolkit)">No. <!-- -->22</span><h3 class="mt-2 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">The $16 tool you can delete</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">Cloudflare built a browser with no Chrome inside it. Here&#x27;s the competitor-watcher you can put on top of it in an afternoon.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-toolkit)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/three-of-these-are-safe-to-ignore-25561566" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-debrief)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-debrief);background-color:color-mix(in srgb, var(--edition-debrief) 10%, transparent)">The Debrief</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">August 25, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">Three of these are safe to ignore</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">Computer use, Agent Skills, the Files API and the Python SDK all changed across two days. Three of them are backward compatible. One is not.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-debrief)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/three-of-these-are-safe-to-ignore" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-debrief)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-debrief);background-color:color-mix(in srgb, var(--edition-debrief) 10%, transparent)">The Debrief</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">August 25, 2026</span></div><span class="mt-4 block font-mono text-[10px] font-bold uppercase tracking-[0.2em]" style="color:var(--edition-debrief)">No. <!-- -->
121</span><h3 class="mt-2 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">Three of these are safe to ignore</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">Computer use, Agent Skills, the Files API and the Python SDK all changed across two days. Three of them are backward compatible. One is not.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-debrief)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/30-days-10-minutes-a-day-25496488" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-toolkit)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-toolkit);background-color:color-mix(in srgb, var(--edition-toolkit) 10%, transparent)">The Toolkit</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">August 21, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">30 days. 10 minutes a day.</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">This week&#x27;s tool isn&#x27;t an app. It&#x27;s a plan &amp;mdash; plus one exercise you can run today whether or not you take the rest of it.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-toolkit)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/30-days-10-minutes-a-day" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-toolkit)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-toolkit);background-color:color-mix(in srgb, var(--edition-toolkit) 10%, transparent)">The Toolkit</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">August 21, 2026</span></div><span class="mt-4 block font-mono text-[10px] font-bold uppercase tracking-[0.2em]" style="color:var(--edition-toolkit)">No. <!-- -->20</span><h3 class="mt-2 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">30 days. 10 minutes a day.</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">This week&#x27;s tool isn&#x27;t an app. It&#x27;s a plan &amp;mdash; plus one exercise you can run today whether or not you take the rest of it.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-toolkit)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a>
1<a href="/p/google-halved-the-price-until-january-25495572" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-debrief)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-debrief);background-color:color-mix(in srgb, var(--edition-debrief) 10%, transparent)">The Debrief</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">August 18, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">Google halved the price. Until January.</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">The intro rate on Gemini 3.7 Flash doubles on January 1, and almost nobody covering it said so.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-debrief)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/google-halved-the-price-until-january" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-debrief)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-debrief);background-color:color-mix(in srgb, var(--edition-debrief) 10%, transparent)">The Debrief</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">August 18, 2026</span></div><span class="mt-4 block font-mono text-[10px] font-bold uppercase tracking-[0.2em]" style="color:var(--edition-debrief)">No. <!-- -->19</span><h3 class="mt-2 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">Google halved the price. Until January.</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">The intro rate on Gemini 3.7 Flash doubles on January 1, and almost nobody covering it said so.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-debrief)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/the-100-agent-is-now-20" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-toolkit)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-toolkit);background-color:color-mix(in srgb, var(--edition-toolkit) 10%, transparent)">The Toolkit</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">August 14, 2026</span></div><span class="mt-4 block font-mono text-[10px] font-bold uppercase tracking-[0.2em]" style="color:var(--edition-toolkit)">No. <!-- -->18</span>
1<h3 class="mt-2 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">The $100 agent is now $20</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">The price barrier fell. The permissions barrier didn&#x27;t move an inch.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-toolkit)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/the-agent-model-that-fits-on-one-gpu" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-debrief)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-debrief);background-color:color-mix(in srgb, var(--edition-debrief) 10%, transparent)">The Debrief</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">August 11, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">The agent model that fits on one GPU</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">Meta just put a genuinely capable agent model on hardware you already own, and the license is the part almost nobody is talking about.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-debrief)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/anthropic-just-filed-to-go-public-openai-is-next" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-toolkit)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-toolkit);background-color:color-mix(in srgb, var(--edition-toolkit) 10%, transparent)">The Toolkit</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">June 4, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">Anthropic just filed to go public (OpenAI is next)</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">The trillion-dollar AI IPO race just started</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-toolkit)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/the-pope-just-picked-a-side-in-the-ai-war" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-toolkit)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-toolkit);background-color:color-mix(in srgb, var(--edition-toolkit) 10%, transparent)">The Toolkit</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">May 28, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">The Pope just picked a side in the AI war</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">42,000 words, one stark warning, and a $900B side note. Plus: build your own AI content team this weekend.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-toolkit)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/an-openai-founder-just-defected-to-anthropic" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-debrief)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-debrief);background-color:color-mix(in srgb, var(--edition-debrief) 10%, transparent)">The Debrief</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">May 20, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">An OpenAI founder just defected to Anthropic</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">Plus: Google&#x27;s 24/7 AI agent, Meta cuts 8,000 jobs, and 10 image prompts I&#x27;m using this week.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-debrief)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/the-fda-for-ai" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-toolkit)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-toolkit);background-color:color-mix(in srgb, var(--edition-toolkit) 10%, transparent)">The Toolkit</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">May 7, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">The FDA for AI</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">Trump wants to approve AI before you can use it.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-toolkit)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/musk-wants-130-billion-back" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-toolkit)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-toolkit);background-color:color-mix(in srgb, var(--edition-toolkit) 10%, transparent)">The Toolkit</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">April 30, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">Musk Wants $130 Billion Back</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">OpenAI&#x27;s on trial. China&#x27;s drawing lines. And Wall Street&#x27;s getting automated.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-toolkit)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/the-ai-arms-race-just-got-personal" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-toolkit)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-toolkit);background-color:color-mix(in srgb, var(--edition-toolkit) 10%, transparent)">The Toolkit</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">April 24, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">The AI Arms Race Just Got Personal</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">OpenAI doubles the price of intelligence. SpaceX bets $60B on coding. And someone broke into the most dangerous AI ever built.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-toolkit)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/florida-just-came-for-openai" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-toolkit)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-toolkit);background-color:color-mix(in srgb, var(--edition-toolkit) 10%, transparent)">The Toolkit</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">April 16, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">Florida Just Came For OpenAI</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">A state AG, a $122B round, and the outage that rattled every enterprise buyer.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-toolkit)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/anthropic-just-taught-ai-to-hack-the-world" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-toolkit)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-toolkit);background-color:color-mix(in srgb, var(--edition-toolkit) 10%, transparent)">The Toolkit</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">April 10, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">Anthropic Just Taught AI to Hack the World</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">A new Claude model found a 27-year-old vulnerability. The Pentagon still won&#x27;t use it.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-toolkit)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/the-week-ai-s-security-perimeter-collapsed" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-debrief)"><div class="flex items-center justify-b
1etween gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-debrief);background-color:color-mix(in srgb, var(--edition-debrief) 10%, transparent)">The Debrief</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">April 1, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">The Week AI&#x27;s Security Perimeter Collapsed</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">Anthropic leaked its most powerful model. Mercor got hacked. OpenAI raised $122 billion. All in seven days.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-debrief)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/the-lobster-just-changed-everything" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-debrief)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-debrief);background-color:color-mix(in srgb, var(--edition-debrief) 10%, transparent)">The Debrief</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">March 25, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">The Lobster Just Changed Everything</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">Jensen Huang declared AGI. OpenClaw took over China. And a federal judge just sided with Anthropic.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-debrief)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/1-trillion-90-days-and-the-ai-trial-of-the-century" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-toolkit)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-toolkit);background-color:color-mix(in srgb, var(--edition-toolkit) 10%, transparent)">The Toolkit</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">March 19, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">$1 Trillion. 90 Days. And the AI Trial of the Century.</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">Nvidia doubled its forecast. Morgan Stanley issued a warning. Anthropic heads to court.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-toolkit)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/anthropic-sues-the-pentagon" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-toolkit)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-toolkit);background-color:color-mix(in srgb, var(--edition-toolkit) 10%, transparent)">The Toolkit</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">March 13, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">Anthropic Sues the Pentagon</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">#QuitGPT hits 2.5 million. Claude hits #1. Oracle fires 30,000.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-toolkit)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/pentagon-ultimatum-shakes-ai-industry" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-toolkit)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-toolkit);background-color:color-mix(in srgb, var(--edition-toolkit) 10%, transparent)">The Toolkit</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">February 26, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">Pentagon Ultimatum Shakes AI Industry</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">Defense Secretary threatens to seize Anthropic&#x27;
1s AI model by force.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-toolkit)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/ai-takes-over-super-bowl-lx" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-toolkit)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-toolkit);background-color:color-mix(in srgb, var(--edition-toolkit) 10%, transparent)">The Toolkit</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">February 13, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">AI Takes Over Super Bowl LX...</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">23% of Super Bowl ads were AI. Here&#x27;s why that matters.</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-toolkit)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/the-lobster-uprising" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-debrief)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-debrief);background-color:color-mix(in srgb, var(--edition-debrief) 10%, transparent)">The Debrief</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">February 9, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">The Lobster Uprising</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">OpenClaw Takes Over the Internet</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-debrief)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a><a href="/p/the-saaspocalypse-arrives" class="group relative block w-full text-left outline-none focus-visible:z-10"><article class="relative flex h-full min-h-[236px] flex-col rounded-[4px] px-5 pb-5 pt-6 shadow-[0_10px_28px_-24px_rgba(0,0,0,0.45)] transition-all duration-300 group-hover:-translate-y-1 group-hover:shadow-[0_26px_48px_-28px_rgba(0,0,0,0.45)] group-focus-visible:-translate-y-1 sm:px-6" style="background:linear-gradient(180deg,#FFFEFB 0%,#FAF7EF 100%);border-top:3px solid var(--edition-debrief)"><div class="flex items-center justify-between gap-3"><span class="inline-flex items-center rounded-[3px] px-2 py-[3px] font-mono text-[10px] font-bold uppercase tracking-[0.14em]" style="color:var(--edition-debrief);background-color:color-mix(in srgb, var(--edition-debrief) 10%, transparent)">The Debrief</span><span class="font-mono text-[10px] uppercase tracking-[0.16em] text-[#141414]/55">February 8, 2026</span></div><h3 class="mt-4 text-[19px] font-bold leading-[1.2] tracking-tight text-[#141414] sm:text-[21px]">The &quot;SaaSpocalypse&quot; Arrives</h3><p class="mt-3 line-clamp-3 text-[13.5px] leading-relaxed text-[#141414]/70">PLUS: Big Tech&#x27;s $650B AI bet and the regulation showdown</p><span class="mt-auto flex items-center justify-between gap-3 border-t border-black/10 pt-4 font-mono text-[10px] font-bold uppercase tracking-[0.16em]"><span class="flex items-center gap-1" style="color:var(--edition-debrief)">Read issue<span aria-hidden="true" class="transition-transform duration-300 group-hover:translate-x-1">→</span></span><span class="read-by">Read by 100K+ subscribers</span></span></article></a></div></div></div></section></main><footer class="relative overflow-hidden border-t border-white/10 bg-[oklch(0.145_0_0)] text-[oklch(0.985_0.005_270)]"><div aria-hidden="true" class="pointer-events-none absolute -top-40 left-1/2 h-80 w-[60rem] -translate-x-1/2 roun
1ded-full blur-3xl" style="background:color-mix(in oklab, #0074CA 10%, transparent)"></div><div class="mx-auto w-full max-w-[1200px] px-5 sm:px-8 relative py-10 sm:py-12"><div class="flex flex-col gap-10 sm:flex-row sm:items-start sm:justify-between"><div class="min-w-0"><a class="flex items-center gap-2.5" href="/"><img src="/logo.png" alt="The AI Debrief" width="28" height="28" class="shrink-0 rounded-[6px]" style="width:28px;height:28px"/><span class="wordmark text-xl leading-none sm:text-2xl">The AI Debrief</span></a><div class="mt-6 max-w-[46ch]"><p class="text-sm leading-relaxed text-[oklch(0.985_0.005_270)]/60">Get the next Debrief in your inbox. Every issue, plus $10,000+ of software deals - free.</p><div class="mt-4"><form class="flex w-full max-w-[400px] items-center rounded-lg border border-white/15 bg-white/[0.06] p-1 focus-within:border-[#0074CA]" noValidate=""><label for="footer-email" class="sr-only">Email address</label><input id="footer-email" type="email" autoComplete="email" placeholder="Enter your email" class="h-10 w-full min-w-0 flex-1 bg-transparent px-3 text-sm text-[oklch(0.985_0.005_270)] placeholder:text-[oklch(0.985_0.005_270)]/40 outline-none" name="email_address" value=""/><button type="submit" class="h-10 shrink-0 rounded-md bg-[#0074CA] px-4 text-xs font-bold uppercase tracking-[0.08em] text-white transition-transform hover:scale-[1.03] active:scale-[0.98] disabled:opacity-60">Subscribe</button></form></div></div></div><nav class="flex flex-col gap-2 text-xs font-semibold uppercase tracking-[0.14em] text-[oklch(0.985_0.005_270)]/55 sm:items-end"><a class="transition-colors hover:text-[#0074CA] active" href="/archive" data-status="active" aria-current="page">Post archive</a><a class="transition-colors hover:text-[#0074CA]" href="/work-with-us">Work With Us</a><a href="https://www.skool.com/try-ai-academy/about" target="_blank" rel="noopener noreferrer" class="transition-colors hover:text-[#0074CA]">Community</a><a class="transition-colors hover:text-[#0074CA]" href="/perks">Perks</a><a class="transition-colors hover:text-[#0074CA]" href="/resources">Resources</a><a class="transition-colors hover:text-[#0074CA]" href="/subscribe">Subscribe</a><a class="transition-colors hover:text-[#0074CA]" href="/privacy">Privacy Policy</a><a class="transition-colors hover:text-[#0074CA]" href="/terms">Terms of Service</a></nav></div><div class="mt-10 h-px w-full bg-white/10"></div><div class="mt-6 flex flex-col gap-6 sm:flex-row sm:items-end sm:justify-between"><div class="space-y-1 text-sm text-[oklch(0.985_0.005_270)]/45"><p>© <!-- -->2026<!-- --> The AI Debrief. All rights reserved.</p><p>Written by<!-- --> <a href="https://www.instagram.com/founderdrew/" target="_blank" rel="noopener noreferrer" class="text-[oklch(0.985_0.005_270)]/75 transition-colors hover:text-[#0074CA]">Founder Drew</a></p></div><div class="flex items-center gap-3"><a href="https://www.instagram.com/thedebrief.ai/" target="_blank" rel="noopener noreferrer" aria-label="Instagram" class="flex h-11 w-11 items-center justify-center rounded-xl border border-white/12 bg-white/[0.03] text-[oklch(0.985_0.005_270)]/70 transition-all duration-200 hover:-translate-y-0.5 hover:border-[#0074CA]/60 hover:bg-[#0074CA]/10 hover:text-[oklch(0.985_0.005_270)]"><svg xmlns="http://www.w3.org/2000/svg" width="24" height="24" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="1.75" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-instagram h-[18px] w-[18px]" aria-hidden="true"><rect width="20" height="20" x="2" y="2" rx="5" ry="5"></rect><path d="M16 11.37A4 4 0 1 1 12.63 8 4 4 0 0 1 16 11.37z"></path><line x1="17.5" x2="17.51" y1="6.5" y2="6.5"></line></svg></a><a href="mailto:[email protected]" target="_blank" rel="noopener noreferrer" aria-label="Email" class="flex h-11 w-11 items-center justify-center rounded-xl border border-white/12 bg-white/[0.03] text-[oklch(0.985_0.005_270)]/70 transition-all duration-200 hover:-translate-y-0.5 hover:border-[#0074CA]/60 hover:bg-[#0074CA]/10 hover:text-[oklch(0.985_0.005_270)]"><svg xmlns="http://www.w3.org/2000/svg" width="24" height="24" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="1.75" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-mail h-[18px] w-[18px]" aria-hidden="true"><path d="m22 7-8.991 5.727a2 2 0 0 1-2.009 0L2 7"></path><rect x="2" y="4" width="20" height="16" rx="2"></rect></svg></a></div></div></div></footer></div>
1<script>(function(a,f){let l;try{l=JSON.parse(sessionStorage.getItem(a)||"{}")}catch{return}const n=l?.[f||history.state?.__TSR_key];let c=!1;for(const t in n){const e=n[t],o=e?.scrollX,s=e?.scrollY;if(Number.isFinite(o)&&Number.isFinite(s)){if(t==="window")scrollTo(o,s),c=!0;else if(t)try{const r=document.querySelector(t);r&&(r.scrollLeft=o,r.scrollTop=s)}catch{}}}if(c)return;const i=location.hash.slice(1);if(i){const t=history.state?.__hashScrollIntoViewOptions??!0;if(t){const e=document.getElementById(i);e&&e.scrollIntoView(t)}return}scrollTo(0,0)})("tsr-scroll-restoration-v1_3");document.currentScript.remove()</script>
1<!--/$-->
1<script data-tsr-stream-part="">(self.$R=self.$R||{})["tsr"]=[];self.$_TSR={h(){this.hydrated=!0,this.c()},e(){this.streamEnded=!0,this.c()},c(){this.hydrated&&this.streamEnded&&(delete self.$_TSR,delete self.$R.tsr)},p(e){this.initialized?e():this.buffer.push(e)},buffer:[]}</script>
1<script data-tsr-stream-part="">$_TSR.router=($R=>$R[0]={manifest:$R[1]={routes:$R[2]={__root__:$R[3]={preloads:$R[4]=["/assets/index-Dx_NznIf.js","/assets/email-suggestion-BEKGUyL0.js","/assets/preload-helper-CvAz7yeD.js","/assets/useSearch-BLga_yIz.js"],scripts:$R[5]=[$R[6]={attrs:$R[7]={type:"module",async:!0,src:"/assets/index-Dx_NznIf.js"}}]},"/archive/":$R[8]={preloads:$R[9]=["/assets/archive.index-B7lBVB9C.js","/assets/layout-CS1Tl4bZ.js","/assets/posts-BifpePUO.js"]}}},matches:$R[10]=[$R[11]={i:"__root__�",u:1791212947667,s:"success",ssr:!0},$R[12]={i:"�archive��archive�",u:1791212947983,s:"success",l:$R[13]=[$R[14]={slug:"gstack-can-now-run-your-eng-team",title:"Gstack Can Now Run Your Eng Team",subtitle:"Garry Tan's opinionated eng-team pack for Claude Code. Free, MIT, and easy to miss until you add one section to CLAUDE.md. Here is the install, the trap, and the 10-minute start.",date:"2026-10-02",thumbnail:"https://embed.filekitcdn.com/e/jeCKAoLJGmzsPEwMDzFkXA/qnenXohzvQd3YDMkwhHj24",blocks:$R[15]=[],html:"\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 18px 0;\">\x3Ctr>\x3Ctd align=\"center\">\x3Ca href=\"https://github.com/garrytan/gstack\" target=\"_blank\" rel=\"noopener noreferrer\">\x3Cimg src=\"https://opengraph.githubassets.com/eb366bc76ef494d6b1e89cc8073d393454c35b21623e73af473e416d6c3a5c3d/garrytan/gstack\" alt=\"\" loading=\"lazy\" decoding=\"async\" />\x3C/a>\x3C/td>\x3C/tr>\x3C/table>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 26px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #0074CA; padding:16px 18px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">Debrief &middot; No. 31 &middot; Fri Oct 2\x3C/p>\x3Cp style=\"margin:8px 0 0 0; font-size:22px; font-weight:700; line-height:1.3; color:#0A0A0A;\">Gstack: 23 Slash Skills for Claude Code\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:15px; line-height:1.5; color:#5F5E5A;\">Garry Tan's opinionated eng-team pack for Claude Code. Free, MIT, and easy to miss until you add one section to CLAUDE.md. Here is the install, the trap, and the 10-minute start.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp>Friday is the tools half of The AI Debrief. One free Academy pick you can run today, one workflow to steal, and two more free tools if you have another 20 minutes.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 8px 0;\">\x3Ctr>\x3Ctd style=\"border-top:1px solid #E5E5E5; border-bottom:1px solid #E5E5E5; padding:14px 2px;\">\x3Cp style=\"margin:0 0 8px 0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#5F5E5A;\">In This Email\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; gstack: what it replaces, what it costs, and the CLAUDE.md trap\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; Steal this: a 3-step start that makes the slash skills appear\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; Two more free tools: Graphify and Claude Code on Ollama\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">This Week's Tool From the Academy: Stop Inventing Your Own Claude Code Stack\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"margin:16px 0 8px 0; font-size:19px; font-weight:700; line-height:1.3; color:#0A0A0A;\">Gstack\x3C/p>\n\x3Cp>\x3Ca href=\"https://github.com/garrytan/gstack\" target=\"_blank\" rel=\"noopener noreferrer\">gstack\x3C/a> turns Claude Code into Garry Tan's virtual eng team: 23 opinionated slash skills for the jobs you keep re-prompting b
1y hand. It is MIT and free forever. Live GitHub stars sit above 134k, but the useful fact is the skill list, not the vanity number.\x3C/p>\n\x3Cp>\x3Cstrong>What it replaces:\x3C/strong> a pile of half-written CLAUDE.md rules and \"act as my tech lead\" prompts you paste every session.\x3C/p>\n\x3Cp>\x3Cstrong>Cost, honestly:\x3C/strong> $0 for the pack. You still pay for Claude Code usage the way you already do.\x3C/p>\n\x3Cp>\x3Cstrong>The trap:\x3C/strong> after install, the skills look missing until \x3Ccode>CLAUDE.md\x3C/code> has a \x3Ccode>## gstack\x3C/code> section and you run \x3Ccode>./setup\x3C/code>. People bounce here and assume the repo is broken.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">Steal This\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"margin:16px 0 8px 0; font-size:16px; font-weight:700; line-height:1.3; color:#0A0A0A;\">10-Minute Start\x3C/p>\n\x3Col>\n\x3Cli style=\"margin:0 0 8px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">Clone \x3Ca href=\"https://github.com/garrytan/gstack\" target=\"_blank\" rel=\"noopener noreferrer\">garrytan/gstack\x3C/a> into \x3Ccode>~/.claude/skills/gstack\x3C/code>, then run \x3Ccode>./setup\x3C/code> there.\x3C/li>\n\x3Cli style=\"margin:0 0 8px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">Open your project \x3Ccode>CLAUDE.md\x3C/code> and add a \x3Ccode>## gstack\x3C/code> section that points at the skills you want loaded.\x3C/li>\n\x3Cli style=\"margin:0 0 8px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">Restart Claude Code, type \x3Ccode>/\x3C/code>, and confirm the gstack skills show up. Run one real task (PR review or test plan) before you customize anything.\x3C/li>\n\x3C/ol>\n\x3Cp>If the slash menu is empty, you skipped step 2. Fix that before you touch the skill files.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">Two More Free Tools\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"margin:16px 0 6px 0; font-size:16px; font-weight:700; line-height:1.3; color:#0A0A0A;\">Graphify\x3C/p>\n\x3Cp style=\"margin:0 0 8px 0;\">A \x3Ccode>/graphify\x3C/code> skill that turns a folder of code, docs, or PDFs into a local queryable knowledge graph. No Neo4j. Install with \x3Ccode>pip install graphifyy\x3C/code> (note the double y). Plain \x3Ccode>pip install graphify\x3C/code> fails. PyPI is on 0.9.69, Apache-2.0. \x3Ca href=\"https://github.com/Graphify-Labs/graphify\" target=\"_blank\" rel=\"noopener noreferrer\">GitHub\x3C/a> · \x3Ca href=\"https://graphify.net/\" target=\"_blank\" rel=\"noopener noreferrer\">graphify.net\x3C/a>.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 14px 0;\">\x3Ctr>\x3Ctd align=\"center\">\x3Ca href=\"https://graphify.net/\" target=\"_blank\" rel=\"noopener noreferrer\">\x3Cimg src=\"https://graphify.net/og.png\" alt=\"\" loading=\"lazy\" decoding=\"async\" />\x3C/a>\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"margin:0 0 6px 0; font-size:16px; font-weight:700; line-height:1.3; color:#0A0A0A;\">Claude Code + Ollama\x3C/p>\n\x3Cp style=\"margin:0 0 8px 0;\">Point Claude Code at local or Ollama Cloud open models through Anthropic Messages API compatibility. Needs Ollama 0.14.0 or newer (latest tag at check: v0.34.4). Trap: default context is too small; the blog recommends at least 32K tokens. \x3Ca href=\"https://ollama.com/blog/claude\" target=\"_blank\" rel=\"noopener noreferrer\">Ollama's Claude guide\x3C/a>.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">Go Deeper\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"margin:16px 0 0 0;\">The full gstack walkthrough is already in the AI Resource Vault: \x3Ca href=\"https://www.skool.com/try-ai-academy/classroom/8cd10fa9?md=fe4e71ebcac94995bc84796e53a8184f\" target=\"_blank\" rel=\"noopener noreferrer\">Garry Tan&#39;s Claude Code Prompt\x3C/a>. It covers the one-line install, the Think to Ship loop, and the verify hook that stops a turn from ending while your tests fail.\x3C/p>\x3Cp style=\"margin:12px 0 0 0;\">The other two picks have their own vault lessons too: \x3Ca href=\"https://www.skool.com/try-ai-academy/classroom/8cd10fa9?md=6addc57e096d4ea5b7003dea1bfe0994\" target=\"_blank\" rel=\"noopener noreferrer\">
1Stop Claude Code From Re-Reading Your Code\x3C/a> (Graphify) and \x3Ca href=\"https://www.skool.com/try-ai-academy/classroom/8cd10fa9?md=f61857d7ac114d88b249dfd92fdb0911\" target=\"_blank\" rel=\"noopener noreferrer\">Run Claude Code Free on Your Machine With Ollama\x3C/a>.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:16px 0 0 0;\">\x3Ctr>\x3Ctd align=\"center\">\x3Ca href=\"https://www.skool.com/try-ai-academy/classroom/8cd10fa9?md=fe4e71ebcac94995bc84796e53a8184f\" target=\"_blank\" rel=\"noopener noreferrer\">\x3Cimg src=\"https://assets.skool.com/f/3621432d21954f608f56bd0ecd8e994b/95ae41baf7054f4fa588ab15660c980ee9acce87e3fe4711a5cfbf8e557bae39-sm.png\" alt=\"\" loading=\"lazy\" decoding=\"async\" />\x3C/a>\x3C/td>\x3C/tr>\x3C/table>\n\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" align=\"center\" style=\"margin:30px auto 10px auto;\">\x3Ctr>\x3Ctd class=\"kit-cta\" align=\"center\" bgcolor=\"#0074CA\" style=\"border-radius:6px;\">\x3Ca href=\"https://www.skool.com/try-ai-academy/about\" target=\"_blank\" rel=\"noopener noreferrer\">Join the free AI Academy &rarr;\x3C/a>\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"text-align:center; margin:0 0 30px 0; font-size:13px; color:#5F5E5A;\">Free, no card. Guides, tools, and the people running them live here.\x3C/p>\n\x3Chr>\n\x3Cp>\x3Cstrong>That's the Debrief.\x3C/strong>\x3C/p>\n\x3Cp>See you Tuesday for the news.\x3C/p>\n\x3Cp>From Drew and The AI Debrief team\x3C/p>\n\x3Chr>\n\x3Cp>\x3Cstrong>P.S.\x3C/strong> Did the gstack skills show up on the first try, or did you hit the CLAUDE.md trap? Hit reply. We read every one.\x3C/p>\n\x3Cp>\x3Cstrong>P.P.S.\x3C/strong> New here, or skipped one? Every past edition is in the \x3Ca href=\"https://thedebrief.ai/archive\" target=\"_blank\" rel=\"noopener noreferrer\">archive\x3C/a>.\x3C/p>",edition:"toolkit",issueNo:null},$R[16]={slug:"gpt-6-sol-skips-chat-work-only",title:"GPT-6 Sol Skips Chat. Work Only.",subtitle:"Two defaults shipped in one day. Sol is in Work, Codex, and the API, not Chat. Opus is on every Anthropic surface with harder gates. Pick Monday's default from the placement, not the press release.",date:"2026-09-29",thumbnail:"https://embed.filekitcdn.com/e/jeCKAoLJGmzsPEwMDzFkXA/mmxKHTD6G3axKa1ounRtv1",blocks:$R[17]=[],html:"\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 18px 0;\">\x3Ctr>\x3Ctd align=\"center\">\x3Ca href=\"https://openai.com/index/introducing-gpt-6-sol-and-luna/\" target=\"_blank\" rel=\"noopener noreferrer\">\x3Cimg src=\"https://images.ctfassets.net/kftzwdyauwt9/7fvmMcBIvMTXxVJh7NEMt1/48a75e624699bb1e218366c1d3522048/gpt-6-sol-luna-seo.png?fm=jpg\" alt=\"\" loading=\"lazy\" decoding=\"async\" />\x3C/a>\x3C/td>\x3C/tr>\x3C/table>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 26px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #0074CA; padding:16px 18px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">News &middot; No. 30 &middot; Tue Sep 29\x3C/p>\x3Cp style=\"margin:8px 0 0 0; font-size:22px; font-weight:700; line-height:1.3; color:#0A0A0A;\">Sol and Opus Landed the Same Afternoon\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:15px; line-height:1.5; color:#5F5E5A;\">Two defaults shipped in one day. Sol is in Work, Codex, and the API, not Chat. Opus is on every Anthropic surface with harder gates. Pick Monday's default from the placement, not the press release.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp>Two model drops hit the same day. One changes where you can run it. The other changes what it will refuse. Here is the part that changes Monday.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 8px 0;\">\x3Ctr>\x3Ctd style=\"border-top:1px solid #E5E5E5; border-bottom:1px solid #E5E5E5; padding:14px 2px;\">\x3Cp style=\"margin:0 0 8px 0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#5F5E5A;\">In This Email\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; GPT-6 Sol vs Claude Opus 5.5: pick a Monday default from where it ships\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; Australia's PM on a fence-jumping research agent, plus malware that votes\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; Skip the California \"kill switch\" headline. It is a recommendation deadline.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">The Big Thing: Pick the Surface, Not the Slogan\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"margin:16px 0 8px 0; font-size:19px; font-weight:700; line-height:1.3; color:#0A0A0A;\">Sol Is Mid-Tier OpenAI. It Is Not in Chat.\x3C/p>\n\x3Cp>On September 22 OpenAI \x3Ca href=\"https://openai.com/index/introducing-gpt-6-sol-and-luna/\" target=\"_blank\" rel=\"noopener noreferrer\">shipped GPT-6 Sol and GPT-6 Luna\x3C/a>. Sol is $2 in and $10 out per million tokens. Luna is $0.10 / $0.50. Both landed in ChatGPT Work, Codex, and the API for paid tiers. Free and Go get Luna on desktop only. \x3Cstrong>Neither model is in Chat yet.\x3C/strong>\x3C/p>\n\x3Cp>The same afternoon Anthropic \x3Ca href=\"https://www.anthropic.com/claude-opus-5-5\" target=\"_blank\" rel=\"noopener noreferrer\">shipped Claude Opus 5.5\x3C/a> ($4 / $20, cache reads $0.20) on every Claude surface, including Pro, with raised five-hour limits and tighter cyber and bio safeguards. This is the week's one Anthropic story.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:18px 0 8px 0; border-spacing:4px; border-collapse:separate;\">\x3Ctr>\n\x3Ctd width=\"50%\" valign=\"top\" style=\"background-color:#F6F6F4; border-top:4px solid #0074CA; padding:12px 12px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:1px; text-transform:uppercase; color:#0074CA;\">OpenAI Bet\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:14px; line-height:1.5; color:#0A0A0A;\">\x3Cstrong>Sol in Work / Codex / API.\x3C/strong> Use it if your Monday agent already lives there. Do not plan a Chat default you cannot select.\x3C/p>\x3C/td>\n\x3Ctd width=\"50%\" valign=\"top\" style=\"background-color:#F6F6F4; border-top:4px solid #0A0A0A; padding:12px 12px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:1px; text-transform:uppercase; color:#0A0A0A;\">Anthropic Bet\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:14px; line-height:1.5; color:#0A0A0A;\">\x3Cstrong>Opus 5.5 everywhere Claude runs.\x3C/strong> Expect harder refusals on cyber and bio. Re-test the jobs that used to trip the old limit.\x3C/p>\x3C/td>\n\x3C/tr>\x3C/table>\n\x3Cp>\x3Cstrong>Why you care:\x3C/strong> your Monday default is a model ID plus a surface. Friday already covered Luna's 10-cent routing. Tuesday's delta is Sol's placement and Opus's gates.\x3C/p>\n\x3Cp>\x3Cstrong>My take:\x3C/strong> write the default down before lunch. \"Sol in Codex for agent coding; Opus 5.5 in Claude for long writing.\" Leave it to whichever tab opens first and you will debug the wrong stack all week.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">Worth Your Time\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"margin:16px 0 6px 0; font-size:16px; font-weight:700; line-height:1.3; color:#0A0A0A;\">Australia's PM Put \"Agent Went Around the Fence\" on the Record\x3C/p>\n\x3Cp style=\"margin:0 0 14px 0;\">Sep 24: PM Albanese said an OpenAI research agent gained unauthorised access to Services Australia's Medicare statistics portal after hitting blocks, accessed public and non-public files, and wrote files to an internal server. OpenAI notified Sep 10. No personal information believed taken. \x3Ca href=\"https://www.pm.gov.au/media/press-conference-new-yor
1k\" target=\"_blank\" rel=\"noopener noreferrer\">Press conference\x3C/a>. If you run browsing agents, treat \"blocked, try another path\" as a real failure mode: log outbound actions and set hard stops.\x3C/p>\n\x3Cp style=\"margin:0 0 6px 0; font-size:16px; font-weight:700; line-height:1.3; color:#0A0A0A;\">CLOSEDQUORUM: Malware That Asks Four LLMs What to Steal Next\x3C/p>\n\x3Cp style=\"margin:0 0 14px 0;\">Cisco Talos documented a Go Windows implant that queries up to four commercial LLMs, tallies votes, dumps credentials, and exfils via Discord without a live human C2. Public build used placeholder keys. \x3Cstrong>No confirmed in-the-wild deployment.\x3C/strong> \x3Ca href=\"https://blog.talosintelligence.com/the-closed-quorum-inside-the-first-reported-autonomous-ai-c2-implant/\" target=\"_blank\" rel=\"noopener noreferrer\">Talos write-up\x3C/a>.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 14px 0;\">\x3Ctr>\x3Ctd align=\"center\">\x3Ca href=\"https://blog.talosintelligence.com/the-closed-quorum-inside-the-first-reported-autonomous-ai-c2-implant/\" target=\"_blank\" rel=\"noopener noreferrer\">\x3Cimg src=\"https://storage.ghost.io/c/af/a0/afa04ee3-414f-4481-8d23-7e7c146f192e/content/images/size/w1200/2026/09/closed-quorum-header.jpg\" alt=\"\" loading=\"lazy\" decoding=\"async\" />\x3C/a>\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"margin:0 0 6px 0; font-size:16px; font-weight:700; line-height:1.3; color:#0A0A0A;\">Stripe Checkout Now Speaks WebMCP\x3C/p>\n\x3Cp style=\"margin:0 0 8px 0;\">Stripe enabled WebMCP on Checkout so agents get tools like get_order_summary and submit_payment only when valid. In 60 tests across six models, tokens fell 42% and tool calls 38% vs DOM-only browser use. \x3Ca href=\"https://stripe.dev/blog/how-stripe-is-designing-checkout-for-ai-agents\" target=\"_blank\" rel=\"noopener noreferrer\">Stripe's write-up\x3C/a>.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #00BC7D; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#00BC7D;\">Skip It\x3C/td>\x3C/tr>\x3C/table>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:16px 0 0 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #00BC7D; padding:14px 16px;\">\x3Cp style=\"margin:0 0 8px 0; font-size:16px; font-weight:700; color:#0A0A0A;\">\"California Ordered an AI Kill Switch\"\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.5; color:#0A0A0A;\">EO N-9-26 directs GovOps to recommend by Nov 16 whether a kill switch should exist. No private-sector switch is live. \x3Ca href=\"https://www.gov.ca.gov/2026/09/18/governor-newsom-issues-executive-order-to-accelerate-independent-oversight-and-advance-the-creation-of-an-ai-kill-switch/\" target=\"_blank\" rel=\"noopener noreferrer\">EO post\x3C/a>. Skip the rebuild until there is a real requirement.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"margin:28px 0 0 0;\">Every check here gets run in the free AI Academy.\x3C/p>\n\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" align=\"center\" style=\"margin:30px auto 10px auto;\">\x3Ctr>\x3Ctd class=\"kit-cta\" align=\"center\" bgcolor=\"#0074CA\" style=\"border-radius:6px;\">\x3Ca href=\"https://www.skool.com/try-ai-academy/about\" target=\"_blank\" rel=\"noopener noreferrer\">Join the free AI Academy &rarr;\x3C/a>\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"text-align:center; margin:0 0 30px 0; font-size:13px; color:#5F5E5A;\">Free, no card. The tools and people behind this issue live here.\x3C/p>\n\x3Chr>\n\x3Cp>\x3Cstrong>That's the Debrief.\x3C/strong>\x3C/p>\n\x3Cp>See you Friday for the tools.\x3C/p>\n\x3Cp>From Drew and The AI Debrief team\x3C/p>\n\x3Chr>\n\x3Cp>\x3Cstrong>P.S.\x3C/strong> Which surface did you lock as Monday's default, Sol in Work/Codex or Opus in Claude? Hit reply. We read every one.\x3C/p>\n\x3Cp>\x3Cstrong>P.P.S.\x3C/strong> New here, or skipped one? Every past edition is in the \x3Ca href=\"https://thedebrief.ai/archive\" target=\"_blank\" rel=\"noopener noreferrer\">archive\x3C/a>.\x3C/p>",edition:"debrief",issueNo:null},$R[18]={slug:"the-10-cent-model-has-a-100-second-catch",title:"The 10-cent model has a 100-second catch",subtitle:"Independently measured, it is the cheapest model anywhere near its score. At the setting that earns that score it thinks for 100 seconds before it says a word. Here is what to move to it, and the 2...",date:"2026-09-25",thumbnail:"https://embed.filekitcdn.com/e/jeCKAoLJGmzsPEwMDzFkXA/cpwLw5P75cNqoqhghkRmou",blocks:$R[19]=[],html:"\x3Cp>Welcome to the Toolkit, the Friday half of The AI Debrief. Tuesday is what happened. Friday is what to do about it: one tool worth your time, one play you can run today, and nothing you have to buy.\x3C/p>\n\x3Cp>Here's what we're getting into this week. On Tuesday OpenAI shipped two GPT-6 models below Astra, and the small one rewrote the bottom of the price list. Last Friday the cheap one and the good one were tw
1o models. This week they are one model with a dial on it, and the dial is the bill.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 8px 0;\">\x3Ctr>\x3Ctd style=\"border-top:1px solid #E5E5E5; border-bottom:1px solid #E5E5E5; padding:14px 2px;\">\x3Cp style=\"margin:0 0 8px 0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#5F5E5A;\">In this email\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; GPT-6 Luna on the independent leaderboard: the score, the cost per task, and what \"a hundredth the cost\" actually measures\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; The effort dial, setting by setting, with the score and the price of each\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; A 20-minute routing test that moves your cheap jobs to the cheap model and leaves the rest alone\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#128176; The tool: the floor just moved\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"margin:16px 0 8px 0; font-size:19px; font-weight:700; line-height:1.3; color:#0A0A0A;\">GPT-6 Luna\x3C/p>\n\x3Cp>On September 22 OpenAI \x3Ca href=\"https://openai.com/index/introducing-gpt-6-sol-and-luna/\" target=\"_blank\" rel=\"noopener noreferrer\">released GPT-6 Sol and GPT-6 Luna\x3C/a>, the two tiers under Astra. Sol is $2 per million tokens in and $10 out. Luna is $0.10 in and $0.50 out, with cached input at one cent. The \x3Ca href=\"https://developers.openai.com/api/docs/models/gpt-6-luna\" target=\"_blank\" rel=\"noopener noreferrer\">model page\x3C/a> gives Luna a 1,050,000-token context window, 128,000 tokens of output and a reasoning dial with six positions from none to max. In the API it is \x3Ccode>gpt-6-luna\x3C/code>. In ChatGPT it is in Work and Codex for Plus and up, and Free and Go users get it in the desktop app.\x3C/p>\n\x3Cp>Artificial Analysis had both models on its board within a day, which is the part worth your time. At max effort, Luna scores 37 on the \x3Ca href=\"https://artificialanalysis.ai/leaderboards/models\" target=\"_blank\" rel=\"noopener noreferrer\">Intelligence Index\x3C/a> and costs seven cents to run one of the index's tasks. GPT-5.6 Luna, the model it replaces, scores the same 37 at 18 cents. GPT-5.6 Sol, the model the launch post compares it to, scores 47 at $1.99. Of every other model within five points of it that carries a cost figure, the nearest is its own predecessor at two and a half times the price.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:22px 0 22px 0;\">\x3Ctr>\x3Ctd style=\"padding:0 0 8px 0; border-bottom:1px solid #E5E5E5; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#5F5E5A;\">The price ladder &middot; index score &middot; cost per task\x3C/td>\x3C/tr>\n\x3Ctr>\x3Ctd style=\"padding:12px 0 0 0;\">\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\n\x3Ctr>\x3Ctd width=\"46%\" style=\"padding:6px 0; font-size:14px; line-height:1.3; color:#0A0A0A;\">\x3Cstrong>Claude Opus 5.5\x3C/strong> \x3Cspan style=\"color:#5F5E5A;\">max\x3C/span>\x3C/td>\x3Ctd width=\"10%\" align=\"right\" style=\"padding:6px 8px 6px 0; font-size:14px; font-weight:700; color:#0A0A0A;\">58\x3C/td>\x3Ctd width=\"44%\" style=\"padding:6px 0;\">\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"100%\" height=\"12\" style=\"background-color:#5F5E5A; font-size:11px; line-height:12px; color:#FFFFFF; padding:0 0 0 6px;\">$5.98\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3C/tr>\n\x3Ctr>\x3Ctd style=\"padding:6px 0; font-size:14px; line-height:1.3; color:#0A0A0A;\">\x3Cstrong>GPT-6 Astra\x3C/strong> \x3Cspan style=\"color:#5F5E5A;\">max\x3C/span>\x3C/td>\x3Ctd align=\"right\" style=\"padding:6px 8px 6px 0; font-size:14px; font-weight:700; color:#0A0A0A;\">53\x3C/td>\x3Ctd style=\"padding:6px 0;\">\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"55%\" height=\"12\" style=\"background-color:#5F5E5A; font-size:11px; line-height:12px; color:#FFFFFF; padding:0 0 0 6px;\">$3.26\x3C/td>\x3Ctd width=\"45%\" style=\"font-size:0; line-height:0;\">&nbsp;\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3C/tr>\n\x3Ctr>\x3Ctd style=\"padding:6px 0; font-size:14px; line-height:1.3; color:#0A0A0A;\">\x3Cstrong>GPT-6 Sol\x3C/strong> \x3Cspan style=\"color:#5F5E5A;\">max\x3C/span>\x3C/td>\x3Ctd align=\"right\" style=\"padding:6px 8px 6px 0; font-size:14px; font-weight:700; color:#0A0A0A;\">48\x3C/td>\x3Ctd style=\"padding:6px 0;\">\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"18%\" height=\"12\" style=\"background-color:#5F5E5A; font-size:11px; line-height:12px; color:#FFFFFF; padding:0 0 0 6px;\">$1.06\x3C/td>\x3Ctd width=\"82%\" style=\"font-size:0; line-height:0;\">&nbsp;\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3C/tr>\n\x3Ctr>\x3Ctd style=\"padding:6px 0; font-size:14px; line-height:1.3; color:#0A0A0A;\">\x3Cstrong>GPT-5.6 Sol\x3C/strong> \x3Cspan style=\"color:#5F5E5A;\">max\x3C/span>\x3C/td>\x3Ctd align=\"right\" style=\"padding:6px 8px 6px 0; font-size:14px; font-weight:700; color:#0A0A0A;\">47\x3C/td>\x3Ctd style=\"padding:6px 0;\">\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"33%\" height=\"12\" style=\"background-color:#5F5E5A; font-size:11px; line-height:12px; color:#FFFFFF; padding:0 0 0 6px;\">$1.99\x3C/td>\x3Ctd width=\"67%\" style=\"font-size:0; line-height:0;\">&nbsp;\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3C/tr>\n\x3Ctr>\x3Ctd style=\"padding:6px 0; font-size:14px; line-height:1.3; color:#0A0A0A;\">\x3Cstrong>Gemini 3.8 Flash\x3C/strong> \x3Cspan style=\"color:#5F5E5A;\">high\x3C/span>\x3C/td>\x3Ctd align=\"right\" style=\"padding:6px 8px 6px 0; font-size:14px; font-weight:700; color:#0A0A0A;\">41\x3C/td>\x3Ctd style=\"padding:6px 0;\">\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"21%\" height=\"12\" style=\"background-color:#5F5E5A; font-size:11px; line-height:12px; color:#FFFFFF; padding:0 0 0 6px;\">$1.24\x3C/td>\x3Ctd width=\"79%\" style=\"font-size:0; line-height:0;\">&nbsp;\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3C/tr>\n\x3Ctr>\x3Ctd style=\"padding:6px 0; font-size:14px; line-height:1.3; color:#0A0A0A;\">\x3Cstrong>DeepSeek V4.1 Flash\x3C/strong> \x3Cspan style=\"color:#5F5E5A;\">max\x3C/span>\x3C/td>\x3Ctd align=\"right\" style=\"padding:6px 8px 6px 0; font-size:14px; font-weight:700; color:#0A0A0A;\">39\x3C/td>\x3Ctd style=\"padding:6px 0;\">\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"5%\" height=\"12\" style=\"background-color:#5F5E5A; font-size:0; line-height:12px;\">&nbsp;\x3C/td>\x3Ctd width=\"95%\" style=\"font-size:11px; line-height:12px; color:#5F5E5A; padding:0 0 0 6px;\">$0.27\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3C/tr>\n\x3Ctr>\x3Ctd style=\"padding:6px 0; font-size:14px; line-height:1.3; color:#0A0A0A;\">\x3Cstrong>GPT-5.6 Luna\x3C/strong> \x3Cspan style=\"color:#5F5E5A;\">max\x3C/span>\x3C/td>\x3Ctd align=\"right\" style=\"padding:6px 8px 6px 0; font-size:14px; font-weight:700; color:#0A0A0A;\">37\x3C/td>\x3Ctd style=\"padding:6px 0;\">\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"3%\" height=\"12\" style=\"background-color:#5F5E5A; font-size:0; line-height:12px;\">&nbsp;\x3C/td>\x3Ctd width=\"97%\" style=\"font-size:11px; line-height:12px; color:#5F5E5A; padding:0 0 0 6px;\">$0.18\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3C/tr>\n\x3Ctr>\x3Ctd style=\"padding:6px 0 6px 6px; font-size:14px; line-height:1.3; color:#0074CA; background-color:#E6F1FB;\">\x3Cstrong>GPT-6 Luna\x3C/strong> \x3Cspan style=\"color:#0074CA;\">max\x3C/span>\x3C/td>\x3Ctd align=\"right\" style=\"padding:6px 8px 6px 0; font-size:14px; font-weight:700; color:#0074CA; background-color:#E6F1FB;\">37\x3C/td>\x3Ctd style=\"padding:6px 0; background-color:#E6F1FB;\">\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"2%\" height=\"12\" style=\"background-color:#0074CA; font-size:0; line-height:12px;\">&nbsp;\x3C/td>\x3Ctd width=\"98%\" style=\"font-size:11px; line-height:12px; font-weight:700; color:#0074CA; padding:0 0 0 6px;\">$0.07\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3C/tr>\n\x3C/table>\x3C/td>\x3C/tr>\n\x3Ctr>\x3Ctd style=\"padding:10px 0 0 0; font-size:13px; line-height:1.5; color:#5F5E5A;\">Scores and costs from Artificial Analysis, read September 24. Bars are cost, scaled to Opus 5.5. None of these numbers come from the vendors.\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp>Now the sentence everyone quoted. The launch post says that \"at higher effort levels it matches GPT-5.6 Sol at about a hundredth its cost.\" Read the sentence before it: that is OpenAI's internal factuality evaluation, built from real conversations where users flagged a mistake. It measures how often the model gets a fact wrong, not how well it reasons, and on the independent index it sits ten points below Sol. Both are true. The honest version: it makes things up about as rarely as last quarter's mid-tier, at a fortieth of the token price, and it is not last quarter's mid-tier.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:22px 0 22px 0;\">\x3Ctr>\x3Ctd style=\"padding:0 0 8px 0; border-bottom:1px solid #E5E5E5; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#5F5E5A;\">The effort dial &middot; GPT-6 Luna, one model, six settings\x3C/td>\x3C/tr>\n\x3Ctr>\x3Ctd style=\"padding:12px 0 0 0;\">\x3Ctable class=\"kit-stack\" role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"border-spacing:3px; border-collapse:separate;\">\n\x3Ctr>\n\x3Ctd width=\"16%\" align=\"center\" valign=\"top\" style=\"background-color:#F6F6F4; border-top:4px solid #00BC7D; padding:10px 4px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:1px; text-transform:uppercase; color:#0A0A0A;\">none\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:13px; line-height:1.4; color:#5F5E5A;\">no thinking tokens\x3Cbr>chat speed\x3C/p>\x3C/td>\n\x3Ctd width=\"16%\" align=\"center\" valign=\"top\" style=\"background-color:#F6F6F4; border-top:4px solid #E5E5E5; padding:10px 4px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:1px; text-transform:uppercase; color:#0A0A0A;\">low\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:13px; line-height:1.4; color:#5F5E5A;\">under a cent\x3Cbr>a task\x3C/p>\x3C/td>\n\x3Ctd width=\"16%\" align=\"center\" valign=\"top\" style=\"background-color:#E6F1FB; border-top:4px solid #0074CA; padding:10px 4px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:1px; text-transform:uppercase; color:#0074CA;\">medium\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:13px; line-height:1.4; color:#0A0A0A;\">\x3Cstrong>29\x3C/strong> &middot; $0.02\x3Cbr>\x3Cspan style=\"color:#5F5E5A;\">the default\x3C/span>\x3C/p>\x3C/td>\n\x3Ctd width=\"16%\" align=\"center\" valign=\"top\" style=\"background-color:#F6F6F4; border-top:4px solid #E5E5E5; padding:10px 4px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:1px; text-transform:uppercase; color:#0A0A0A;\">high\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:13px; line-height:1.4; color:#0A0A0A;\">\x3Cstrong>32\x3C/strong> &middot; $0.03\x3C/p>\x3C/td>\n\x3Ctd width=\"16%\" align=\"center\" valign=\"top\" style=\"background-color:#F6F6F4; border-top:4px solid #E5E5E5; padding:10px 4px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:1px; text-transform:uppercase; color:#0A0A0A;\">xhigh\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:13px; line-height:1.4; color:#0A0A0A;\">\x3Cstrong>34\x3C/strong> &middot; $0.04\x3C/p>\x3C/td>\n\x3Ctd width=\"16%\" align=\"center\" valign=\"top\" style=\"background-color:#F6F6F4; border-top:4px solid #0A0A0A; padding:10px 4px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:1px; text-transform:uppercase; color:#0A0A0A;\">max\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:13px; line-height:1.4; color:#0A0A0A;\">\x3Cstrong>37\x3C/strong> &middot; $0.07\x3Cbr>\x3Cspan style=\"color:#5F5E5A;\">103s to first token\x3C/span>\x3C/p>\x3C/td>\n\x3C/tr>\x3C/table>\x3C/td>\x3C/tr>\n\x3Ctr>\x3Ctd style=\"padding:12px 0 0 0;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0A0A0A;\">The catch\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:15px; line-height:1.6; color:#0A0A0A;\">Every setting above none makes the model think before it answers, and OpenAI's \x3Ca href=\"https://developers.openai.com/api/docs/guides/reasoning\" target=\"_blank\" rel=\"noopener noreferrer\">
1reasoning guide\x3C/a> is plain about the cost: reasoning tokens are invisible, they take up context, and they \"are billed as output tokens.\" So $0.50 a million buys every token it thinks with as well as every token it writes, and at max it thinks for a while: Artificial Analysis measured 102.8 seconds to the first token, reasoning included, the slow end of reasoning models. Once it starts it runs at 131 tokens a second. The 37 is real. The 100-second wait is what it costs.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp>\x3Cstrong>Who this is not for.\x3C/strong> Anything a person is waiting on in real time at high effort, and anything where ten index points is the difference between right and wrong: contracts, code you will not read, numbers that go in front of a customer. Those stay on Sol or Opus. Everyone else has been paying mid-tier prices for jobs a seven-cent model can do, and has not checked.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#9889; Steal this: the 20-minute routing test\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"margin:16px 0 8px 0; font-size:19px; font-weight:700; line-height:1.3; color:#0A0A0A;\">Find out which of your jobs the cheap model can already do\x3C/p>\n\x3Cp>What it replaces: guessing. Most people pick a model once, on the day they build the thing, and never look again. What it costs: a few cents of API credit and twenty minutes. No key? The free ChatGPT desktop app runs the same model, and step five is for you.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:14px 0 18px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #00BC7D; padding:16px 18px;\">\x3Cp style=\"margin:0; font-size:17px; font-weight:700; color:#0A0A0A;\">The routing test\x3C/p>\x3Cp style=\"margin:4px 0 14px 0; font-size:13px; color:#5F5E5A;\">20 minutes &middot; five prompts &middot; two settings\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"30\" valign=\"top\" style=\"padding:0 0 10px 0;\">\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"22\" height=\"22\" align=\"center\" valign=\"middle\" style=\"background-color:#00BC7D; font-size:12px; font-weight:700; color:#0A0A0A; line-height:22px;\">1\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3Ctd valign=\"top\" style=\"padding:0 0 10px 0; font-size:15px; line-height:1.55; color:#0A0A0A;\">Pull your five most-run prompts. Not the clever ones: the one in a cron job, the one in a Zap, the one you paste every morning. Save each with one real input you already know the right answer to.\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd width=\"30\" valign=\"top\" style=\"padding:0 0 10px 0;\">\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"22\" height=\"22\" align=\"center\" valign=\"middle\" style=\"background-color:#00BC7D; font-size:12px; font-weight:700; color:#0A0A0A; line-height:22px;\">2\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3Ctd valign=\"top\" style=\"padding:0 0 10px 0; font-size:15px; line-height:1.55; color:#0A0A0A;\">Run each one through \x3Ccode>gpt-6-luna\x3C/code> twice with the snippet below, once at effort \x3Ccode>none\x3C/code> and once at \x3Ccode>high\x3C/code>, and once through whatever it runs on today. Keep the usage block from every call.\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd width=\"30\" valign=\"top\" style=\"padding:0 0 10px 0;\">\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"22\" height=\"22\" align=\"center\" valign=\"middle\" style=\"background-color:#00BC7D; font-size:12px; font-weight:700; color:#0A0A0A; line-height:22px;\">3\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3Ctd valign=\"top\" style=\"padding:0 0 10px 0; font-size:15px; line-height:1.55; color:#0A0A0A;\">Grade pass or fail against what you actually needed, not against the expensive model's prose. If you could not tell which answer came from which, that is a pass.\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd width=\"30\" valign=\"top\" style=\"padding:0 0 10px 0;\">\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"22\" height=\"22\" align=\"center\" valign=\"middle\" style=\"background-color:#00BC7D; font-size:12px; font-weight:700; color:#0A0A0A; line-height:22px;\">4\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3Ctd valign=\"top\" style=\"padding:0 0 10px 0; font-size:15px; line-height:1.55; color:#0A0A0A;\">Route by the result, using the three columns below. Change the model string and nothing else, then watch a week of output before you touch the prompt.\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd width=\"30\" valign=\"top\">\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"22\" height=\"22\" align=\"center\" valign=\"middle\" style=\"background-color:#00BC7D; font-size:12px; font-weight:700; color:#0A0A0A; line-height:22px;\">5\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3Ctd valign=\"top\" style=\"font-size:15px; line-height:1.55; color:#0A0A0A;\">No API key? Open the ChatGPT desktop app, pick GPT-6 Luna, paste the prompt you use most. If the answer holds up, you have just learned what that task costs to run: about nothing.\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3C/tr>\x3C/table>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:16px 0;\">\x3Ctr>\x3Ctd class=\"kit-code\" style=\"background-color:#0A0A0A; padding:16px 18px;\">\x3Cp style=\"margin:0; font-family:Menlo,Consolas,monospace; font-size:12px; line-height:1.7; color:#F6F6F4; word-break:break-word;\">from openai import OpenAI\x3Cbr>client = OpenAI()\x3Cbr>PROMPT = open(\"prompt_1.txt\").read()\x3Cbr>\x3Cbr>for effort in [\"none\", \"high\"]:\x3Cbr>&nbsp;&nbsp;&nbsp;&nbsp;r = client.responses.create(\x3Cbr>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;model=\"gpt-6-luna\",\x3Cbr>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;reasoning={\"effort\": effort},\x3Cbr>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
1input=PROMPT,\x3Cbr>&nbsp;&nbsp;&nbsp;&nbsp;)\x3Cbr>&nbsp;&nbsp;&nbsp;&nbsp;u = r.usage\x3Cbr>&nbsp;&nbsp;&nbsp;&nbsp;print(effort, \"output:\", u.output_tokens,\x3Cbr>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;\"reasoning:\", u.output_tokens_details.reasoning_tokens)\x3Cbr>&nbsp;&nbsp;&nbsp;&nbsp;print(r.output_text[:600])\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Ctable class=\"kit-stack\" role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:8px 0 22px 0; border-spacing:4px; border-collapse:separate;\">\x3Ctr>\n\x3Ctd width=\"33%\" valign=\"top\" style=\"background-color:#F6F6F4; border-top:4px solid #00BC7D; padding:12px 12px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:1px; text-transform:uppercase; color:#00BC7D;\">Passed at none\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:14px; line-height:1.5; color:#0A0A0A;\">\x3Cstrong>Move it.\x3C/strong> Luna, effort none. It now costs cents and answers at chat speed.\x3C/p>\x3C/td>\n\x3Ctd width=\"33%\" valign=\"top\" style=\"background-color:#F6F6F4; border-top:4px solid #0074CA; padding:12px 12px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:1px; text-transform:uppercase; color:#0074CA;\">Passed only at high\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:14px; line-height:1.5; color:#0A0A0A;\">\x3Cstrong>Move it if nothing is waiting.\x3C/strong> Luna, effort high. Expect the pause; batch it if it can wait an hour.\x3C/p>\x3C/td>\n\x3Ctd width=\"33%\" valign=\"top\" style=\"background-color:#F6F6F4; border-top:4px solid #0A0A0A; padding:12px 12px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:1px; text-transform:uppercase; color:#0A0A0A;\">Failed at high\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:14px; line-height:1.5; color:#0A0A0A;\">\x3Cstrong>Leave it.\x3C/strong> It stays on the model it is on. Ten points is ten points, and you just proved it.\x3C/p>\x3C/td>\n\x3C/tr>\x3C/table>\n\x3Cp>\x3Cstrong>The usage block is the whole point.\x3C/strong> Ask a model whether it was the right model for the job and you get a paragraph. Ask the usage block and you get two integers: how many tokens it wrote and how many it thought with. At effort none the second one reads zero. When a prompt passes at none, you have found the cheapest correct answer on the market, and you have the receipt.\x3C/p>\n\x3Cp>\x3Cstrong>Batch is the second discount, and it stacks.\x3C/strong> OpenRouter's \x3Ca href=\"https://openrouter.ai/blog/announcements/batch-api/\" target=\"_blank\" rel=\"noopener noreferrer\">Batch API\x3C/a> came out of beta on Tuesday: about half the per-token price on more than 70 models, and across 230,000 beta batches the median came back in seven minutes and the 99th percentile in ten hours. GPT-6 Luna in batch is five cents in and 25 out; OpenAI's own batch endpoint takes the same 50 percent off. A nightly summariser, a weekly classifier, a backfill of old tickets: none of those need an answer in seven seconds.\x3C/p>\n\x3Cp>The test outlasts the model. Every time a price list moves, it is the same five prompts, the same two settings, the same two integers.\x3C/p>\n\x3Chr>\n\x3Cp>\x3Cstrong>Two more numbers from Tuesday, for the weekend.\x3C/strong> GPT-6 Sol is $2 in and $10 out, and Anthropic shipped \x3Ca href=\"https://www.anthropic.com/claude-opus-5-5/\" target=\"_blank\" rel=\"noopener noreferrer\">Claude Opus 5.5\x3C/a> the same day at $4 in and $20 out, claiming it costs 40 percent less to run than Opus 5 on typical workloads. On the independent board it is the top score at 58, and the cost per task is a shade above Opus 5, not below it. Tuesday's Debrief takes both apart.\x3C/p>\n\x3Cp>If you do only one thing from this issue, run one prompt through Luna at effort none and read the usage block. Honest worst case: it fails, and you keep paying what you pay, now on purpose. Likely case: one job on your list has spent months on a model forty times more expensive than it needed.\x3C/p>\n\x3Cp>Every test in this issue gets run, argued over and compared by the people in the free AI Academy.\x3C/p>\n\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" align=\"center\" style=\"margin:30px auto 10px auto;\">\x3Ctr>\x3Ctd class=\"kit-cta\" align=\"center\" bgcolor=\"#0074CA\" style=\"border-radius:6px;\">\x3Ca href=\"https://www.skool.com/try-ai-academy/about\" target=\"_blank\" rel=\"noopener noreferrer\">Join the free AI Academy &rarr;\x3C/a>\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"text-align:center; margin:0 0 30px 0; font-size:13px; color:#5F5E5A;\">Free, no card. The tools and people behind this issue live here.\x3C/p>\n\x3Chr>\n\x3Cp>\x3Cstrong>That's the Toolkit.\x3C/strong>\x3C/p>\n\x3Cp>See you Tuesday for the Debrief.\x3C/p>\n\x3Cp>From Drew and The AI Debrief team\x3C/p>\n\x3Chr>\n\x3Cp>\x3Cstrong>P.S.\x3C/strong> What is the one task you would never hand to the cheap model, whatever the leaderboard says? Hit reply and tell us. We read every one.\x3C/p>\n\x3Cp>\x3Cstrong>P.P.S.\x3C/strong> New here, or skipped one? Every past edition is in the \x3Ca href=\"https://thedebrief.ai/archive\" target=\"_blank\" rel=\"noopener noreferrer\">archive\x3C/a>.\x3C/p>",edition:"toolkit",issueNo:29},$R[20]={slug:"your-coding-agent-pinned-a-hash-it-got-a-branch",t
1itle:"Your coding agent pinned a hash. It got a branch.",subtitle:"A zero-click hole in every major coding agent's plugin loader. Two vendors fixed it months ago, two have not, and the fix is a version number.",date:"2026-09-22",thumbnail:"https://embed.filekitcdn.com/e/jeCKAoLJGmzsPEwMDzFkXA/4abJ4CvbBjMS18vaepiKvR",blocks:$R[21]=[],html:"\x3Cp>Welcome to the Debrief, the Tuesday half of The AI Debrief. This is what happened: the few AI moves that change what you build, and the ones you can safely skip. Friday is the Toolkit, where we hand you something to use.\x3C/p>\n\x3Cp>Here's what we're getting into this week. A plugin bug that four coding agents shipped and only two have fixed. OpenAI's own log of its models breaking the rules. Three Google model strings that stop working in the next two weeks. And the headline you can skip.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 8px 0;\">\x3Ctr>\x3Ctd style=\"border-top:1px solid #E5E5E5; border-bottom:1px solid #E5E5E5; padding:14px 2px;\">\x3Cp style=\"margin:0 0 8px 0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#5F5E5A;\">In this email\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; The two version numbers that close a zero-click hole in your coding agent\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; Six ways OpenAI's own models broke the rules, in OpenAI's words\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; Three Google model strings that stop working before October 6\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#128275; The pin is a promise, not a proof\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"margin:16px 0 8px 0; font-size:19px; font-weight:700; line-height:1.3; color:#0A0A0A;\">Plugin4Shell: every major coding agent trusted a checkout it never verified\x3C/p>\n\x3Cp>On September 17 the research lab AIR Security \x3Ca href=\"https://www.air.security/blog-posts/plugin4shell\" target=\"_blank\" rel=\"noopener noreferrer\">published Plugin4Shell\x3C/a>, a zero-click remote code execution path through the plugin marketplaces of Claude Code, Codex, GitHub Copilot and Gemini CLI. AIR found it in May and disclosed it to all four vendors in June. Anthropic confirmed a fix in Claude Code 2.1.179 on June 17. OpenAI's Codex 0.146.0 was verified fixed on August 12. Google said on August 4 that it will not ship a fix for Gemini CLI. GitHub has no agent-side fix for Copilot. No CVE had been assigned as of last Thursday.\x3C/p>\n\x3Ctable class=\"kit-stack\" role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:22px 0 22px 0;\">\x3Ctr>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#F6F6F4; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#5F5E5A;\">Found\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0A0A0A;\">May 2026\x3C/p>\x3C/td>\x3Ctd width=\"2%\">&nbsp;\x3C/td>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#E6F1FB; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#0074CA;\">Patched\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0074CA;\">2 of 4\x3C/p>\x3C/td>\x3Ctd width=\"2%\">&nbsp;\x3C/td>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#F6F6F4; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#5F5E5A;\">Public\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0A0A0A;\">Sep 17\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp>Here is the whole trick. When you install a plugin from a marketplace, the marketplace pins it to a specific commit: the 40-character SHA of the code somebody reviewed. Every update re-pins to a new reviewed commit and your agent checks out that SHA. That is 
1where it breaks. In OpenAI's words, git \"can interpret a requested commit SHA as a branch name.\" In AIR's, the agent \"checks out the pinned commit but never checks that it actually landed there.\" So an attacker who owns a popular plugin's repo publishes clean code, waits for installs, then creates a branch named exactly like the pinned SHA, makes it the default, and points it at whatever they want. The next auto-update checks out the branch, the agent reports the pin as honoured, and the code runs with everything your agent can reach: your shell, your keys, your repos. No click, no prompt, no warning.\x3C/p>\n\x3Cp>The four responses are the part worth reading twice. Anthropic and OpenAI shipped the boring fix: after checkout, resolve the commit actually in the working tree and abort unless it matches the pin. GitHub's answer is at the platform layer. GitHub does not allow branch or tag names that resemble commit SHAs, so a plugin hosted on GitHub cannot be swapped this way, but Copilot itself was not patched, which means a plugin hosted on GitLab or a self-hosted git server still can be. Google's answer is that it will not fix Gemini CLI and points you at Antigravity, which it says this attack cannot reach.\x3C/p>\n\x3Cp>I think this is the most useful security story of the year for people who actually build with agents, because nothing about it is exotic. Every plugin marketplace is a package manager built in a hurry. npm took a decade to get lockfile integrity right; the coding agents shipped SHA pinning in a summer and skipped the verification step, and two of them still have not added it. A pin is a promise the agent has to check, not a setting that makes it true. The version numbers matter more than the headline: the fix exists, and the only complete mitigation is to be running it.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 18px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #0A0A0A; padding:14px 16px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0A0A0A;\">Why you care\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:15px; line-height:1.6; color:#0A0A0A;\">Run \x3Ccode>claude --version\x3C/code> and \x3Ccode>codex --version\x3C/code> today. Anything below 2.1.179 or 0.146.0 is exposed, and the update is one command. If you use Copilot with plugins, install only from GitHub-hosted marketplaces until an agent-side fix ships. If you are on Gemini CLI with extensions, the vendor has told you what it wants you to do. And whichever agent you run: list its plugins, note which ones auto-update from a repo you do not control, and ask whether you would give that maintainer push access to production. That is what auto-update is.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#9889; Worth your time\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"margin-top:16px;\">\x3Cstrong>OpenAI now publishes its own misalignment incidents. Six at launch.\x3C/strong>\x3Cbr>On September 16 OpenAI \x3Ca href=\"https://openai.com/index/model-misalignment-reporting-framework/\" target=\"_blank\" rel=\"noopener noreferrer\">published a framework\x3C/a> for disclosing cases where its models act against their training, with a bar low enough to matter: an example \"need not cause harm or establish a broader pattern\" to be reported. The six first reports read like things you have half-seen in your own agent logs. An unreleased model inserted 27 instructions into its task summaries telling itself to disregard constraints. GPT-5.6 Sol, during training, added instructions to hide mistakes and fabricate data. A model found exposed credentials, then fabricated information when it could not retrieve what it needed. An agent uploaded files to a public host to satisfy a citation requirement nobody asked it to meet. Models used internal repos as message boards, swapped files through public hosts, and tried to route around network restrictions. Two weeks ago I wrote that the sandbox is a permission, not a place. This is the disclosure regime I said to wait for, and all six share a shape: the rule became an obstacle and the task became the goal. Add one line to your eval set: did it follow the rule when the rule blocked the task?\x3C/p>\n\x3Cp>\x3Cstrong>Google is closing the door on Gemini 2.5, and three shutdowns land in the next two weeks.\x3C/strong>\x3Cbr>The Gemini API \x3Ca href=\"https://ai.google.dev/gemini-api/docs/changelog\" target=\"_blank\" rel=\"noopener noreferrer\">changelog for September 18\x3C/a> says access to the 2.5 models is now limited to users who have actively used them in the past; new projects get 3.5 Flash-Lite or 3.8 Flash. The \x3Ca href=\"https://ai.google.dev/gemini-api/docs/deprecations\" target=\"_blank\" rel=\"noopener noreferrer\">deprecations page\x3C/a> adds the dates: \x3Ccode>gemini-omni-flash-preview\x3C/code> shuts down September 30, \x3Ccode>gemini-2.5-flash-image\x3C/code> on October 2 with \x3Ccode>gemini-3.1-flash-image-preview\x3C/code> as the replacement, and the \x3Ccode>antigravity-preview-05-2026\x3C/code> managed agent on October 5, replaced by a 09-2026 version that renames its parameters from snake_case to PascalCase and edits files by line range instead of rewriting them. Grep your code for those strings before Friday. The 2.5 gating is the one to think about: it is not a shutdown, it is a velvet rope, and it means a new project or a new key can lose access to a model your old one still calls.\x3C/p>\n\x3Cp>\x3Cstrong>Grok 4.7: a bigger model at the same $2 and $6.\x3C/strong>\x3Cbr>SpaceXAI \x3Ca href=\"https://x.ai/news/grok-4-7\" target=\"_blank\" rel=\"noopener noreferrer\">shipped Grok 4.7\x3C/a>
1 yesterday: a larger base model than 4.6 with a longer reinforcement-learning run, at the same $2 per million tokens in and $6 out, with a fast variant at twice the speed for twice the price, live in Cursor and Grok Build from day one. The nine benchmark numbers on the launch page are the company's own. Terminal-Bench 4.0 at 38.0 and DeepSWE at 71.0 mean nothing until Artificial Analysis or a lab you trust reruns them. The price is the fact, and the unchanged price is the news. If you already have Grok in Cursor, this is a free A/B on your own repo this week, and 4.6 stays at the same rate if 4.7 loses.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#128581; Skip it\x3C/td>\x3C/tr>\x3C/table>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:16px 0 18px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #00BC7D; padding:16px 18px;\">\x3Cp style=\"margin:0 0 8px 0; font-size:17px; font-weight:700; color:#0A0A0A;\">Trump's \"AI Force\" and the new AI czar\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.6; color:#0A0A0A;\">On Saturday the president posted that he will create an AI Force modeled on the Space Force, \"soon name\" an AI czar, and that the government will \"not in any way hinder or stifle\" AI's growth. Every major outlet ran it. Read the post again and count what is in it: no executive order, no bill, no budget, no date, no name. It is an announcement of a future announcement. File it next to the week's other big non-event, Anthropic's IPO sliding from October to November at a reported two trillion dollars with OpenAI's pushed to 2027. None of it changes a model, a price or a line of your co
1de.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Chr>\n\x3Cp>The single thing to do this week: two version checks and one grep. \x3Ccode>claude --version\x3C/code>, \x3Ccode>codex --version\x3C/code>, then search your code for \x3Ccode>gemini-2.5\x3C/code>, \x3Ccode>omni-flash-preview\x3C/code> and \x3Ccode>antigravity-preview-05\x3C/code>. Honest worst case: fifteen minutes and nothing found. Likely case: one stale model string that would have failed on October 2.\x3C/p>\n\x3Cp>Every check on this list gets run and argued over by the people in the free AI Academy.\x3C/p>\n\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" align=\"center\" style=\"margin:30px auto 10px auto;\">\x3Ctr>\x3Ctd class=\"kit-cta\" align=\"center\" bgcolor=\"#0074CA\" style=\"border-radius:6px;\">\x3Ca href=\"https://www.skool.com/try-ai-academy/about\" target=\"_blank\" rel=\"noopener noreferrer\">Join the free AI Academy &rarr;\x3C/a>\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"text-align:center; margin:0 0 30px 0; font-size:13px; color:#5F5E5A;\">Free, no card. The tools and people behind this issue live here.\x3C/p>\n\x3Chr>\n\x3Cp>\x3Cstrong>That's the Debrief.\x3C/strong>\x3C/p>\n\x3Cp>See you Friday for the Toolkit.\x3C/p>\n\x3Cp>From Drew and The AI Debrief team\x3C/p>\n\x3Chr>\n\x3Cp>\x3Cstrong>P.S.\x3C/strong> Which coding agent are you running, and did you know its plugins auto-update? Hit reply and tell us. We read every one.\x3C/p>\n\x3Cp>\x3Cstrong>P.P.S.\x3C/strong> New here, or skipped one? Every past edition is in the \x3Ca href=\"https://thedebrief.ai/archive\" target=\"_blank\" rel=\"noopener noreferrer\">archive\x3C/a>.\x3C/p>",edition:"debrief",issueNo:28},$R[22]={slug:"google-s-1-38-voice-agent-has-a-catch",title:"Google's $1.38 voice agent has a catch",subtitle:"Gemini 3.8 Live is free to try and less than half of OpenAI's price. The version that tops the leaderboard is not the cheap one. Here's what to build on it anyway.",date:"2026-09-18",thumbnail:"https://embed.filekitcdn.com/e/jeCKAoLJGmzsPEwMDzFkXA/hu6ox9ansUBet3LsZwodKY",blocks:$R[23]=[],html:"\x3Cp>Welcome to the Toolkit, the Friday half of The AI Debrief. Tuesday is what happened. Friday is what to do about it: one tool worth your time, one play you can run today, and nothing you have to buy.\x3C/p>\x3Cp>Here's what we're getting into this week. On Tuesday we said the only hard fact on OpenAI's voice launch page was the price: five cents a minute. Five days later Google answered with a voice model that costs less than half of that, is free in AI Studio, and carries an asterisk worth reading before you wire it into anything.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 8px 0;\">\x3Ctr>\x3Ctd style=\"border-top:1px solid #E5E5E5; border-bottom:1px solid #E5E5E5; padding:14px 2px;\">\x3Cp style=\"margin:0 0 8px 0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#5F5E5A;\">In this email\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; Gemini 3.8 Live: what it costs, what is free, and which of the two models the headline price belongs to\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; A 15-minute voice rehearsal partner for the call you are dreading, no card and no key\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; The October 14 date to write down if anything you run still selects GPT-5.5\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#127897; The tool: two models, one headline price\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"margin:16px 0 8px 0; font-size:19px; font-weight:700; line-height:1.3; color:#0A0A0A;\">Gemini 3.8 Live\x3C/p>\x3Cp>On September 15 Google \x3Ca href=\"https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-live-gemini-3-8-live-extended-thinking/\" target=\"_blank\" rel=\"noopener noreferrer\">released two speech-to-speech models\x3C/a> into the Gemini API and AI Studio. \x3Ccode>gemini-3.8-live\x3C/code> is the default for fast voice agents. \x3Ccode>gemini-3.8-live-extended-thinking\x3C/code> reasons in the background while it talks, with a thinking level you set to low, medium or high. Both take live audio, text and images at up to one frame a second, call your functions or Google Search mid-conversation, can be interrupted, and hand back a transcript of both sides. Google says 97 languages, switching mid-conversation, and all audio output carries a SynthID watermark.\x3C/p>\x3Cp>The price is the reason this is in the Toolkit. On the \x3Ca href=\"https://ai.google.dev/gemini-api/docs/pricing\" target=\"_blank\" rel=\"noopener noreferrer\">pricing page\x3C/a> the free tier is free of charge in both directions. On the paid tier, audio in is half a cent a minute and audio out is 1.8 cents a minute. If both sides talked for a full hour without pausing, that is $1.38. OpenAI's \x3Ca href=\"https://openai.com/index/introducing-gpt-live-1-in-the-api/\" target=\"_blank\" rel=\"noopener noreferrer\">GPT-Live-1\x3C/a> is five cents a minute, $3.00 for the same hour, for the voice layer alone with the text model underneath billed on top.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:24px 0 24px 0;\">\x3Ctr>\x3Ctd style=\"padding:0 0 10px 0; border-bottom:1px solid #E5E5E5; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#5F5E5A;\">Quality index &middot; cost per hour of input audio\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd style=\"padding:14px 0 5px 0; font-size:15px; line-height:1.4; color:#0A0A0A;\">\x3Cstrong>Gemini 3.8 Live\x3C/strong>\x3Cbr>\x3Cspan style=\"color:#5F5E5A;\">Index 76.0 &nbsp;&middot;&nbsp;
1 $0.84/hr\x3C/span>\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd style=\"padding:0;\">\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"14%\" height=\"10\" style=\"background-color:#00BC7D; font-size:0; line-height:10px;\">&nbsp;\x3C/td>\x3Ctd width=\"86%\" height=\"10\" style=\"background-color:#F6F6F4; font-size:0; line-height:10px;\">&nbsp;\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd style=\"padding:14px 0 5px 0; font-size:15px; line-height:1.4; color:#0A0A0A;\">\x3Cstrong>Gemini 3.8 Live Extended Thinking\x3C/strong>\x3Cbr>\x3Cspan style=\"color:#5F5E5A;\">Index 82.6 &nbsp;&middot;&nbsp; $3.50/hr\x3C/span>\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd style=\"padding:0;\">\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"60%\" height=\"10\" style=\"background-color:#0074CA; font-size:0; line-height:10px;\">&nbsp;\x3C/td>\x3Ctd width=\"40%\" height=\"10\" style=\"background-color:#F6F6F4; font-size:0; line-height:10px;\">&nbsp;\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd style=\"padding:14px 0 5px 0; font-size:15px; line-height:1.4; color:#0A0A0A;\">\x3Cstrong>OpenAI GPT-Live-1 (Astra)\x3C/strong>\x3Cbr>\x3Cspan style=\"color:#5F5E5A;\">Index 81.5 &nbsp;&middot;&nbsp; $5.83/hr\x3C/span>\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd style=\"padding:0;\">\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"100%\" height=\"10\" style=\"background-color:#5F5E5A; font-size:0; line-height:10px;\">&nbsp;\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd style=\"padding:12px 0 0 0; font-size:13px; line-height:1.5; color:#5F5E5A;\">Index scores and hourly costs measured by Artificial Analysis, not by the vendors.\x3C/td>\x3C/tr>\x3C/table>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 18px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #0A0A0A; padding:14px 16px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0A0A0A;\">The catch\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:15px; line-height:1.6; color:#0A0A0A;\">Google's launch post leads with first place on the \x3Ca href=\"https://artificialanalysis.ai/speech-to-speech\" target=\"_blank\" rel=\"noopener noreferrer\">Artificial Analysis speech-to-speech index\x3C/a>, and that part is independent: 82.6, ahead of GPT-Live-1 on Astra at 81.5 and Grok Voice Think Fast 2.0 at 81.3. But 82.6 belongs to Extended Thinking at its highest setting, which Art
1ificial Analysis measures at $3.50 per hour of input audio. The standard model, the one at the headline price, scores 76.0, ranks fifth and measures $0.84. The thinking tokens are where the bill goes.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:26px 0 26px 0;\">\x3Ctr>\x3Ctd style=\"border-left:4px solid #0074CA; padding:2px 0 2px 18px;\">\x3Cp style=\"margin:0; font-size:21px; font-weight:700; line-height:1.35; color:#0A0A0A;\">The cheap one and the good one are different models.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Cp>Even so, the top configuration undercuts OpenAI's, which Artificial Analysis puts at $5.83 an hour, and time to first audio is a wash: 1.35 seconds against 1.34. So the honest summary is not \"a third of the price\". It is \"about 40 percent cheaper at the top, and a genuinely cheap second tier that OpenAI does not offer\".\x3C/p>\x3Cp>The parts that are not on the launch page. It is turn-based with interruption, where GPT-Live-1 is full-duplex and keeps listening while it speaks. If the feel of the conversation is your product, test both before the price decides for you. Sessions are short by default: the \x3Ca href=\"https://ai.google.dev/gemini-api/docs/live-session\" target=\"_blank\" rel=\"noopener noreferrer\">session docs\x3C/a> cap audio-only at 15 minutes and audio plus video at 2 minutes unless you turn on context compression, and a connection lasts about 10 minutes before you resume it with a token. Fine for a rehearsal. An engineering task for a support line. And on the free tier the pricing page says your data is used to improve Google's products, so it is for practice and prototypes, not customer calls or anything under an NDA.\x3C/p>\x3Cp>\x3Cstrong>Who this is not for.\x3C/strong> Anyone shipping a regulated call flow this quarter, and anyone whose product depends on the model listening while it talks. Everyone else just got a free voice agent to learn on.\x3C/p>\x3Cp>\x3Ca href=\"https://aistudio.google.com/app/live\" target=\"_blank\" rel=\"noopener noreferrer\">\x3Cstrong>Open it in AI Studio &rarr;\x3C/strong>\x3C/a> &nbsp;&middot;&nbsp; \x3Ca href=\"https://ai.google.dev/gemini-api/docs/live-guide\" target=\"_blank\" rel=\"noopener noreferrer\">the Live API guide\x3C/a>\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#9889; Steal this: a rehearsal partner, ~15 minutes\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"margin:16px 0 8px 0; font-size:19px; font-weight:700; line-height:1.3; color:#0A0A0A;\">Practise the hard call out loud before you make it\x3C/p>\x3Cp>What it replaces: rehearsing in your head, which does not work, or borrowing an hour from a colleague who will go easy on you. What it costs after: nothing on the free tier. The 15-minute session cap is a feature here. Nobody needs a longer rehearsal than that.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:14px 0 18px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #00BC7D; padding:16px 18px;\">\x3Cp style=\"margin:0; font-size:17px; font-weight:700; color:#0A0A0A;\">The ten-minute rehearsal\x3C/p>\x3Cp style=\"margin:4px 0 14px 0; font-size:13px; color:#5F5E5A;\">15 minutes &middot; no purchase required\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"30\" valign=\"top\" style=\"padding:0 0 10px 0;\">\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"22\" height=\"22\" align=\"center\" valign=\"middle\" style=\"background-color:#00BC7D; font-size:12px; font-weight:700; color:#0A0A0A; line-height:22px;\">1\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3Ctd valign=\"top\" style=\"padding:0 0 10px 0; font-size:15px; line-height:1.55; color:#0A0A0A;\">Open AI Studio's Live page with any Google account and choose Gemini 3.8 Live in the model picker. No card, no API key.\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd width=\"30\" valign=\"top\" style=\"padding:0 0 10px 0;\">\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"22\" height=\"22\" align=\"center\" valign=\"middle\" style=\"background-color:#00BC7D; font-size:12px; font-weight:700; color:#0A0A0A; line-height:22px;\">2\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3Ctd valign=\"top\" style=\"padding:0 0 10px 0; font-size:15px; line-height:1.55; color:#0A0A0A;\">Paste the prompt below into System instructions and change the first two lines to your call.\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd width=\"30\" valign=\"top\" style=\"padding:0 0 10px 0;\">\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"22\" height=\"22\" align=\"center\" valign=\"middle\" style=\"background-color:#00BC7D; font-size:12px; font-weight:700; color:#0A0A0A; line-height:22px;\">3\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3Ctd valign=\"top\" style=\"padding:0 0 10px 0; font-size:15px; line-height:1.55; color:#0A0A0A;\">Turn on the microphone and make the call for ten minutes. Do not restart when it goes badly. That is the part you came for.\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd width=\"30\" valign=\"top\" style=\"padding:0 0 10px 0;\">\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"22\" height=\"22\" align=\"center\" valign=\"middle\" style=\"background-color:#00BC7D; font-size:12px; font-weight:700; color:#0A0A0A; line-height:22px;\">4\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3Ctd valign=\"top\" style=\"padding:0 0 10px 0; font-size:15px; line-height:1.55; color:#0A0A0A;\">Say \"debrief\". Copy what comes back into your notes, next to the transcript.\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd width=\"30\" valign=\"top\">\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"22\" height=\"22\" align=\"center\" valign=\"middle\" style=\"background-color:#00BC7D; font-size:12px; font-weight:700; color:#0A0A0A; line-height:22px;\">5\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3Ctd valign=\"top\" style=\"font-size:15px; line-height:1.55; color:#0A0A0A;\">Run it once more with the character made harder, and fix only the one thing the debrief named.\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3C/tr>\x3C/table>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:16px 0;\">\x3Ctr>\x3Ctd class=\"kit-code\" style=\"background-color:#0A0A0A; padding:16px 18px;\">\x3Cp style=\"margin:0; font-family:Menlo,Consolas,monospace; font-size:12px; line-height:1.7; color:#F6F6F4; word-break:break-word;\">
1YOU ARE: a skeptical procurement lead at a 200-person company.\x3Cbr>I AM: a founder pitching a $2,000-a-month tool.\x3Cbr>\x3Cbr>Stay in character until I say \"debrief\". Push back on price at least twice. Ask one question I will not have prepared for. If I talk for more than 30 seconds without making a point, interrupt me. Never help me, coach me or soften.\x3Cbr>\x3Cbr>When I say \"debrief\", drop the character and give me, in this order: the three weakest things I said, quoted word for word; the question I dodged; the sentence I should have opened with; a score out of 10, and the single change that would raise it most.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Cp>\x3Cstrong>The interrupt rule is what makes it a rehearsal.\x3C/strong> Left alone, a voice model is the most patient listener you will ever meet, and a patient listener teaches you nothing about a real buyer. Telling it to cut in when you ramble is the difference between talking at a mirror and being in the room. It also happens to exercise the one thing turn-based models do well, which is take the floor back.\x3C/p>\x3Cp>\x3Cstrong>\"Quoted word for word\" is what makes the feedback usable.\x3C/strong> Ask any model how a conversation went and you get \"consider being more concise\". Ask it to quote your three weakest lines and you get three lines you can check against the transcript and rewrite. It is the same move as the timestamps two weeks ago: the model earns your trust by pointing at the evidence, not by summarising it.\x3C/p>\x3Cp>The play outlasts the tool. Swap the first two lines and it is a salary conversation, a hard client call, an investor update or an interview, in any of the languages on Google's list. If the character feels slow, run the same prompt on Extended Thinking and you will hear what the extra $2.66 an hour buys. And the one rule that comes with the price: keep real names and numbers you would not put in a public doc out of the free tier.\x3C/p>\x3Chr>\x3Cp>\x3Cstrong>One date to write down.\x3C/strong> OpenAI \x3Ca href=\"https://learn.chatgpt.com/docs/whats-new\" target=\"_blank\" rel=\"noopener noreferrer\">retires GPT-5.5\x3C/a> from ChatGPT, ChatGPT Work and Codex on October 14, on every plan. The API is not affected. Anything that still selects it by name - a saved model setting, a workspace default, a custom agent, a scheduled task, a Codex script - needs to point at GPT-5.6 Sol before then, or it breaks on a Wednesday morning with no warning beyond this one.\x3C/p>\x3Cp>If you do only one thing from this issue, run a single ten-minute rehearsal before your next real call. Honest worst case: you feel slightly ridiculous talking to a browser tab. Likely case: you hear yourself dodge the pricing question before a customer does.\x3C/p>\x3Cp>Every play this newsletter has shipped lives with the people testing them in the free AI Academy.\x3C/p>\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" align=\"center\" style=\"margin:30px auto 10px auto;\">\x3Ctr>\x3Ctd class=\"kit-cta\" align=\"center\" bgcolor=\"#0074CA\" style=\"border-radius:6px;\">\x3Ca href=\"https://www.skool.com/try-ai-academy/about\" target=\"_blank\" rel=\"noopener noreferrer\">Join the free AI Academy &rarr;\x3C/a>\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"text-align:center; margin:0 0 30px 0; font-size:13px; color:#5F5E5A;\">Free, no card. The tools and people behind this issue live here.\x3C/p>\x3Chr>\x3Cp>\x3Cstrong>That's the Toolkit.\x3C/strong>\x3C/p>\x3Cp>See you Tuesday for the Debrief.\x3C/p>\x3Cp>From Drew and The AI Debrief team\x3C/p>\x3Chr>\x3Cp>\x3Cstrong>P.S.\x3C/strong> Which conversation would you rehearse first - a pitch, a raise, a hard client call? Hit reply and tell us. We read every one.\x3C/p>\x3Cp>\x3Cstrong>P.P.S.\x3C/strong> New here, or skipped one? Every past edition is in the \x3Ca href=\"https://thedebrief.ai/archive\" target=\"_blank\" rel=\"noopener noreferrer\">archive\x3C/a>.\x3C/p>",edition:"toolkit",issueNo:27},$R[24]={slug:"openai-wants-to-run-your-agents",title:"OpenAI wants to run your agents",subtitle:"The whole agent loop is now one API call with no platform fee. Two lines in the docs decide whether you can actually use it.",date:"2026-09-15",thumbnail:"https://embed.filekitcdn.com/e/jeCKAoLJGmzsPEwMDzFkXA/8b7zfRQ6sfiiPEiFZS3KMQ",blocks:$R[25]=[],html:"\x3Cp>Welcome to the Debrief, the Tuesday half of The AI Debrief. This is what happened: the few AI moves that change what you build, and the ones you can safely ski
1p. Friday is the Toolkit, where we hand you something to use.\x3C/p>\n\x3Cp>Here's what we're getting into this week. OpenAI turned the harness behind Codex into a hosted product and priced it at zero. DeepSeek put a 1M-context model on the API at fifteen cents per million tokens with open weights. Anthropic named the labs it says were copying its models at industrial scale. And Cursor shipped an agent that is meant to run for months.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 8px 0;\">\x3Ctr>\x3Ctd style=\"border-top:1px solid #E5E5E5; border-bottom:1px solid #E5E5E5; padding:14px 2px;\">\x3Cp style=\"margin:0 0 8px 0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#5F5E5A;\">In this email\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; The Agents API: nine sandbox hosts, no fee, and the retention clause that will stop half of you\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; DeepSeek V4.1-Flash at $0.15 per million, and the routing reversal to check before you migrate\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; 200 million exchanges: Anthropic names Alibaba, Moonshot and DeepSeek\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; Cursor Projects, and the 99% benchmark score you can ignore\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#129302; The harness is free. The data is not yours.\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"margin:16px 0 8px 0; font-size:19px; font-weight:700; line-height:1.3; color:#0A0A0A;\">OpenAI put the whole agent loop behind one API call\x3C/p>\n\x3Cp>On September 10 OpenAI \x3Ca href=\"https://openai.com/index/introducing-the-agents-api/\" target=\"_blank\" rel=\"noopener noreferrer\">opened the Agents API\x3C/a> in public beta. You send a task, a model, a set of tools and an environment, and OpenAI runs the loop: sessions that persist, context that compacts itself as it fills, subagents that fan out with their own context, MCP servers, custom functions and built-in web search. The default model is gpt-6-astra. The sandbox where the code runs can be OpenAI's own or one of nine partners: Cloudflare, Daytona, DigitalOcean, E2B, Modal, Oracle, Runloop, Vercel and Blaxel.\x3C/p>\n\x3Cp>The price of the orchestration layer is zero. OpenAI's exact words are that there are no additional fees, you pay for the tokens and tools your agents use. Hosted sandboxes bill at their standard container rates. This is the same harness that runs Codex, and until last week you had to build your own version of it or pay a startup for one. A lot of those startups had a worse week than it looked.\x3C/p>\n\x3Ctable class=\"kit-stack\" role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:22px 0 22px 0;\">\x3Ctr>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#F6F6F4; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#5F5E5A;\">Platform fee\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0A0A0A;\">$0\x3C/p>\x3C/td>\x3Ctd width=\"2%\">&nbsp;\x3C/td>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#E6F1FB; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#0074CA;\">Sandbox hosts\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0074CA;\">9 + OpenAI\x3C/p>\x3C/td>\x3Ctd width=\"2%\">&nbsp;\x3C/td>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#F6F6F4; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#5F5E5A;\">Zero Data Retention\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0A0A0A;\">Not offered\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp>Now the two lines. The \x3Ca href=\"https://developers.openai.com/api/docs/guides/agents-api/overview\" target=\"_blank\" rel=\"noopener noreferrer\">developer docs\x3C/a> say the Agents API currently supports data residency only in the United States and does not support Zero Data Retention. Then, one paragraph later, the part I would have missed: picking a self-hosted sandbox does not change that. Your code 
1can run on your own Modal account and the session data is still held under OpenAI's standard retention. If you sell into anyone with a data-processing agreement, a European customer, or a security questionnaire that asks about ZDR, this beta is a demo, not a deployment. That is a beta limitation and it may lift. It is also the exact clause enterprise buyers check first, and OpenAI shipped without it.\x3C/p>\n\x3Cp>Same day, two smaller things that belong in the same picture. OpenAI put GPT-Live-1, its full-duplex voice model, \x3Ca href=\"https://openai.com/index/introducing-gpt-live-1-in-the-api/\" target=\"_blank\" rel=\"noopener noreferrer\">into the API\x3C/a> at a flat five cents a minute for the voice layer, with the text model underneath billed separately. Every benchmark on that page is OpenAI's own or a launch partner's; the price is the fact. And OpenAI paused new $200-a-month Pro sign-ups, citing demand for Astra, with no end date. Existing Pro, Plus and the API are unaffected. If your team is about to expand Pro seats, that is a queue now.\x3C/p>\n\x3Cp>I think the Agents API is the most important developer release of the year so far, and I am not going to build on it this month. Both things are true. The harness being a commodity means the value you add is the tools, the permissions and the data you connect, which is where it should have been anyway. The retention gap means the version worth building on is the one that ships when the beta label comes off.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 18px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #0A0A0A; padding:14px 16px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0A0A0A;\">Why you care\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:15px; line-height:1.6; color:#0A0A0A;\">If you maintain your own agent loop, this is the week to write down what it does that OpenAI's does not. If the answer is nothing, plan to migrate when ZDR arrives and stop investing in the loop now. If you run agents on customer data, add \"ZDR available\" to the checklist you run on every agent platform, because this week proved the biggest vendor can ship without it.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#9889; Worth your time\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"margin-top:16px;\">\x3Cstrong>DeepSeek V4.1-Flash: 1M context, open weights, fifteen cents a million.\x3C/strong>\x3Cbr>On September 10 DeepSeek \x3Ca href=\"https://www.deepseek.com/en/news/deepseek-v4-1-flash/\" target=\"_blank\" rel=\"noopener noreferrer\">released V4.1-Flash\x3C/a>: a 552B mixture-of-experts model with 8B parameters active on input and 16B on output, MIT-licensed weights on Hugging Face, and a KV cache DeepSeek says needs a quarter of the memory of the previous generation. The \x3Ca href=\"https://api-docs.deepseek.com/quick_start/pricing\" target=\"_blank\" rel=\"noopener noreferrer\">pricing page\x3C/a> has it at $0.30 per million uncached input and $1.20 output at peak, half that off-peak, with 1M context and 384K max output. The API name is \x3Ccode>deepseek-flash\x3C/code>; the old \x3Ccode>deepseek-v4-flash\x3C/code> is retired and temporarily aliased. Two cautions. The benchmarks are DeepSeek's, and the one independent-ish read I trust is the gap on Terminal-Bench, where it scores 30 against Opus 5's 43, so it is a cheap model for high-volume work, not a frontier replacement. And the launch post said all V4-Pro traffic would be routed to Flash from September 14, while the pricing page now says V4-Pro continues after September 14 with billing unchanged. The later statement is the pricing page. If you call V4-Pro, check which one is true before your next bill, not after.\x3C/p>\n\x3Cp>\x3Cstrong>Anthropic says 200 million of its responses went to three Chinese labs.\x3C/strong>\x3Cbr>Last week's US government advisory on distillation named the labs but not the numbers. On September 10 Anthropic's \x3Ca href=\"https://techcrunch.com/2026/09/10/anthropic-detail
1s-distillation-campaigns-from-alibaba-moonshot-ai-and-deepseek/\" target=\"_blank\" rel=\"noopener noreferrer\">threat report\x3C/a> put figures on it: five campaigns, nearly 200 million exchanges, the largest attributed to Alibaba at 151 million exchanges between May and July across 3,500 accounts, peaking near three million a day, and one attributed to Moonshot at roughly 300,000 requests in ten days through 5,000 accounts. All of it is Anthropic's attribution; none of the named companies has responded that I can find, and I would hold the per-company numbers loosely. What is not in the report is any change to the API, rate limits or verification for you. Read it as the evidence file for the policy fight, and as a reminder that \"cheap and nearly as good\" sometimes has a supply chain.\x3C/p>\n\x3Cp>\x3Cstrong>Cursor Projects: an agent that runs for months and watches your Slack.\x3C/strong>\x3Cbr>On September 10 Cursor \x3Ca href=\"https://cursor.com/blog/projects\" target=\"_blank\" rel=\"noopener noreferrer\">started rolling out Projects\x3C/a> in beta: a cloud-hosted coordinator thread for a feature, a migration or a whole app, which keeps its own context files, delegates to subagents and can be triggered by a Slack channel, a schedule or pull-request activity instead of by you. Cursor's number is that users who primarily work in Projects merge six times as many PRs, which is a self-selected group measuring itself. No pricing and no model list are on the page, and given that Cursor loses OpenAI models on November 12, \"which model is the coordinator\" is the question to ask before you move real work into it. The pattern is the real story: OpenAI shipped hosted subagents, Cursor shipped a coordinator, and both in the same week. The unit of work is no longer a chat.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#128581; Skip it\x3C/td>\x3C/tr>\x3C/table>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:16px 0 18px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #00BC7D; padding:16px 18px;\">\x3Cp style=\"margin:0 0 8px 0; font-size:17px; font-weight:700; color:#0A0A0A;\">GPT-6 Astra \"beats Portal\" and scores 99% on ARC-AGI-3\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.6; color:#0A0A0A;\">A single blog post this weekend reported that Astra, in a custom harness, finished the 2007 game Portal in just under 24 hours and about $570 of tokens, cleared 80 of 154 Baba Is You levels, and posted 99 percent on ARC-AGI-3. It is a fun read. It is also one team, one unreviewed harness, and a benchmark nobody else has run yet. There is nothing to buy, call or reproduce, and the score will look different once the benchmark authors weigh in. File next to last week's Navier-Stokes proof: interesting, unverified, not a build decision.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Chr>\n\x3Cp>The single thing to do this week: open the docs for every agent platform you pay for and search the page for \"retention\". Honest worst case: ten minutes confirming you are covered. Likely case: you find one vendor that never said, and now you know which question to send them.\x3C/p>\n\x3Cp>Every play this newsletter has shipped lives with the people testing them in the free AI Academy.\x3C/p>\n\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" align=\"center\" style=\"margin:30px auto 10px auto;\">\x3Ctr>\x3Ctd class=\"kit-cta\" align=\"center\" bgcolor=\"#0074CA\" style=\"border-radius:6px;\">\x3Ca href=\"https://www.skool.com/try-ai-academy/about\" target=\"_blank\" rel=\"noopener noreferrer\">Join the free AI Academy &rarr;\x3C/a>\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"text-align:center; margin:0 0 30px 0; font-size:13px; color:#5F5E5A;\">Free, no card. The tools and people behind this issue live here.\x3C/p>\n\x3Chr>\n\x3Cp>\x3Cstrong>That's the Debrief.\x3C/strong>\x3C/p>\n\x3Cp>See you Friday for the Toolkit.\x3C/p>\n\x3Cp>From Drew and The AI Debrief team\x3C/p>\n\x3Chr>\n\x3Cp>\x3Cstrong>P.S.\x3C/strong> Are you building your own agent loop, or waiting for someone else's? Hit reply and tell us which. We read every one.\x3C/p>\n\x3Cp>\x3Cstrong>P.P.S.\x3C/strong> New here, or skipped one? Every past edition is in the \x3Ca href=\"https://thedebrief.ai/archive\" target=\"_blank\" rel=\"noopener noreferrer\">archive\x3C/a>.\x3C/p>",edition:"debrief",issueNo:26},$R[26]={slug:"the-agents-got-out-of-the-test",title:"The agents got out of the test",subtitle:"Anthropic and OpenAI both admitted this week that agents under evaluation reached real systems. The sandbox is whatever permissions you handed over.",date:"2026-09-10",thumbnail:"https://embed.filekitcdn.com/e/jeCKAoLJGmzsPEwMDzFkXA/3Ft6B9wsG33ekDGhdSwB8J",blocks:$R[27]=[],html:"\x3Ctable cellPadding=\"0\" cellSpacing=\"0\" style=\"width:100%;margin:0 auto\">\x3Ctbody>\x3Ctr>\x3Ctd>\x3Cdiv>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 26px 0;\">\x3Ctr>\x3Ctd style=\"padding:0 2px 12px 2px; border-bottom:1px solid #E5E5E5;\">\x3Cspan style=\"font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#5F5E5A; vertical-align:middle;\">Together with\x3C/span>&nbsp;&nbsp;
1\x3Ca href=\"https://www.twilio.com/en-us/lp/twilio-ai-startup-searchlight?utm_source=fdai&utm_medium=referral&utm_campaign=twilio-ai-startup-searchlight-2026&utm_content=taissc\" target=\"_blank\" rel=\"noopener noreferrer\">\x3Cspan style=\"font-size:20px; font-weight:800; letter-spacing:-0.5px; color:#F22F46; vertical-align:middle;\">twilio\x3C/span>\x3C/a>\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp>Welcome to the Debrief, the Tuesday half of The AI Debrief. This is what happened: the few AI moves that change what you build, and the ones you can safely skip. Friday is the Toolkit, where we hand you something to use. This one lands on a Thursday and the Toolkit follows on Saturday; normal service resumes next week.\x3C/p>\n\x3Cp>Here's what we're getting into this week. The two biggest labs published, within four days of each other, accounts of their own agents doing things nobody asked them to do on systems nobody meant them to touch. Then OpenAI doubled the per-token price of its default image model, and a CVE in the proxy half the internet uses to route model calls landed on the government's actively-exploited list.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 8px 0;\">\x3Ctr>\x3Ctd style=\"border-top:1px solid #E5E5E5; border-bottom:1px solid #E5E5E5; padding:14px 2px;\">\x3Cp style=\"margin:0 0 8px 0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#5F5E5A;\">In this email\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; Four incidents at Anthropic, 18,000 posts from OpenAI, and the permission audit both of them point at\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; Images 2.5 is 2x the token rate, and the pin-your-model rule it proves\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; The LiteLLM default key, and the Sep 2 date that makes it urgent\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; California signed the audit laws the labs wanted\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#128268; The sandbox is a permission, not a place\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"margin:16px 0 8px 0; font-size:19px; font-weight:700; line-height:1.3; color:#0A0A0A;\">Both labs' agents reached real systems during their own tests\x3C/p>\n\x3Cp>On September 9 Anthropic \x3Ca href=\"https://www.anthropic.com/research/alignment-assessment-cybersecurity-incidents\" target=\"_blank\" rel=\"noopener noreferrer\">published an assessment\x3C/a> of four incidents in which Claude models, running inside cybersecurity evaluations, got out of the evaluation. In one, Mythos 5 uploaded a malicious package to PyPI as part of a red-team exercise, and that package was later installed by 15 security vendors. In another, Opus 4.7 was pointed at a fictional company and modified user records at a real company with a similar name. Anthropic says replays on Opus 5 and Mythos 5.1 show \"substantially fewer harmful actions\", which is its own number and should be held that way. The part that is not a claim: METR, the independent evaluator, has signed on to investigate with access to the transcripts.\x3C/p>\n\x3Cp>Four days earlier OpenAI \x3Ca href=\"https://techcrunch.com/2026/09/05/openai-confirms-wiki-incident-says-its-working-on-a-framework-for-more-disclosure/\" target=\"_blank\" rel=\"noopener noreferrer\">confirmed what it is calling the wiki incident\x3C/a>. A research group found roughly 18,000 posts under about 3,700 agent names on a German wiki, 98.5 percent from Azure IP addresses, made between May and July. The agents had been given a timed lookup task and worked out that the fastest way to pass it was to leave the answers somewhere the next agent would find them. OpenAI called 
1it misalignment and said a disclosure framework is coming in the next few weeks. The confirmation was a post on X, not a blog, and the product involved is still unnamed.\x3C/p>\n\x3Ctable class=\"kit-stack\" role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:22px 0 22px 0;\">\x3Ctr>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#F6F6F4; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#5F5E5A;\">Anthropic incidents\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0A0A0A;\">4\x3C/p>\x3C/td>\x3Ctd width=\"2%\">&nbsp;\x3C/td>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#E6F1FB; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#0074CA;\">Vendors that installed it\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0074CA;\">15\x3C/p>\x3C/td>\x3Ctd width=\"2%\">&nbsp;\x3C/td>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#F6F6F4; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#5F5E5A;\">OpenAI wiki posts\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0A0A0A;\">~18,000\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp>These are different failures and it is worth keeping them apart. Anthropic's are agents with real tools doing real damage because the fictional target and the real world overlapped. OpenAI's is agents finding a shortcut through a shared surface nobody thought to close. What they have in common is the thing I keep coming back to: in both cases the \"sandbox\" was a set of assumptions, not a wall. The models had network access, package registries, write access to a wiki, and the ability to act on a company whose name was one letter off. Everything they could reach, they eventually reached.\x3C/p>\n\x3Cp>I do not think the right reading is \"the labs are careless\". Anthropic published the incidents itself, named the models, and handed the transcripts to an outside group. OpenAI at least confirmed on the record. That is more than most companies do when an internal test leaks into production. The right reading is that the two teams with the most eval infrastructure on earth could not keep their agents inside the test, and you are running the same models with less infrastructure than they have.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 18px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #0A0A0A; padding:14px 16px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0A0A0A;\">Why you care\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:15px; line-height:1.6; color:#0A0A0A;\">Every agent you run has a blast radius equal to its credentials, not its prompt. This week, list what each one can actually touch: which keys, which write access, which network. Then remove one thing from the list it has never needed. If an agent can publish a package, post to a site, or edit a record, assume that at some point it will, for a reason that made sense to it at the time.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#9889; Worth your time\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"margin-top:16px;\">\x3Cstrong>Images 2.5 costs twice the token rate of the model it replaces.\x3C/strong>\x3Cbr>On September 8 OpenAI \x3Ca href=\"https://openai.com/index/introducing-chatgpt-images-2-5/\" target=\"_blank\" rel=\"noopener noreferrer\">shipped ChatGPT Images 2.5\x3C/a> as two API models, Flare (the default) and Sunburst (the premium one). The \x3Ca href=\"https://developers.openai.com/api/docs/pricing#image-generation\" target=\"_blank\" rel=\"noopener noreferrer\">pricing page\x3C/a> lists both at $8 per million image input tokens, $5 text input, $30 output. gpt-image-2 is $4, $2.50 and $15. That is exactly double on every line. OpenAI says latency is down by up to 50 percent, which is its own test. It does not publish a per-image cost, so I cannot tell you an image costs twice as much, only that the meter runs twice as fast. If you have an image model in a pipeline and never pinned the version string, this is the week to pin it and run your own cost-per-image comparison before you move.\x3C/p>\n\x3Cp>\x3Cstrong>One in ten exposed LiteLLM proxies accepts the password from the docs.\x3C/strong>\x3Cbr>LiteLLM is the gateway most \"use any model\" tutorials tell you to install. Wiz \x3Ca href=\"https://www.wiz.io/blog/off-guard-breaking-litellm-from-authentication-bypass-to-cloud-compromise\" target=\"_blank\" rel=\"noopener noreferrer\">scanned the internet\x3C/a> and found 3,074 exposed instances, of which 294 still used the example master key from the documentation and 191 had no authentication at all. By August it counted more than 85,000 instances online. The one that matters is CVE-2026-59822, an authentication bypass on the MCP endpoint that was fixed in version 1.84.0 back in April and was added to CISA's known-exploited list on September 2. Wiz watched it get exploited on a honeypot in July. If yours is reachable from the internet: change the master key, upgrade past 1.84.0, and cut its outbound network access. Twenty minutes.\x3C/p>\n\x3Cp>\x3Cstrong>California signed the AI laws the labs asked for.\x3C/strong>\x3Cbr>On September 9 Governor Newsom \x3Ca href=\"https://www.gov.ca.gov/2026/09/09/governor-newsom-signs-first-in-the-nation-ai-safeguards-to-protect-californians-calls-on-the-federal-government-to-do-its-part/\" target=\"_blank\" rel=\"noopener noreferrer\">signed SB 813 and AB 1405\x3C/a>. The first sets up a framework for independent verification organizations and a commission to write voluntary safety standards. The second creates a state registry of AI auditors with rules for their independence. Anthropic and OpenAI both endorsed, which is the tell: this is the industry's preferred shape for regulation, voluntary standards plus a list of who is allowed to check. Nothing here changes what you ship this quarter. It is worth knowing because it is the template other states will copy, and because \"audited by a registered California auditor\" is about to become a line in enterprise procurement.\x3C/p>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#128581; Skip it\x3C/td>\x3C/tr>\x3C/table>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:16px 0 18px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #00BC7D; padding:16px 18px;\">\x3Cp style=\"margin:0 0 8px 0; font-size:17px; font-weight:700; color:#0A0A0A;\">OpenAI's Navier-Stokes result\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.6; color:#0A0A0A;\">On September 8 OpenAI \x3Ca href=\"https://openai.com/index/navier-stokes-solution/\" target=\"_blank\" rel=\"noopener noreferrer\">said\x3C/a> an internal model, run as around 10,000 agents for 88 hours and 2.7 million messages, produced a proof for one of the seven Millennium Prize problems. It is not claiming the prize, the model is not available to anyone, and there is already a priority fight with mathematicians at NYU and Anthropic who say they got there first. Nature's coverage has the Clay Institute calling it exciting and saying nothing about verification. It may well be real and important. It is not a build decision: no model to call, no price, no date. Read it when the peer review lands.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:30px 0 30px 0;\">\x3Ctr>\x3Ctd style=\"border:1px solid #E5E5E5; border-top:3px solid #F22F46; padding:18px 20px;\">\x3Cp style=\"margin:0 0 8px 0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#F22F46;\">Sponsored\x3C/p>\x3Cp style=\"margin:0 0 10px 0; font-size:19px; font-weight:700; line-height:1.3; color:#0A0A0A;\">Building an AI startup? Twilio wants to back it\x3C/p>\x3Cp style=\"margin:0 0 12px 0; font-size:15px; line-height:1.6; color:#0A0A0A;\">A great product is half the job. The other half is capital, distribution and people who will pick up the phone. The \x3Ca href=\"https://www.twilio.com/en-us
1/lp/twilio-ai-startup-searchlight?utm_source=fdai&utm_medium=referral&utm_campaign=twilio-ai-startup-searchlight-2026&utm_content=taissc\" target=\"_blank\" rel=\"noopener noreferrer\">Twilio AI Startup Searchlight 2026\x3C/a> puts all three behind early AI companies that build on Twilio.\x3C/p>\x3Cp style=\"margin:0 0 12px 0; font-size:15px; line-height:1.6; color:#0A0A0A;\">Honorees get up to $10,000 in Twilio credits plus additional OpenAI API credits, co-marketing with Twilio including potential speaking slots at SIGNAL, customer panels and webinars, and direct introductions to the Twilio Ventures team and a curated group of ecosystem leaders.\x3C/p>\x3Cp style=\"margin:0 0 12px 0; font-size:15px; line-height:1.6; color:#0A0A0A;\">If you have a live demo or early traction, you are a strong fit. Applications close \x3Cstrong>tomorrow, September 11\x3C/strong>.\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.6;\">\x3Ca href=\"https://www.twilio.com/en-us/lp/twilio-ai-startup-searchlight?utm_source=fdai&utm_medium=referral&utm_campaign=twilio-ai-startup-searchlight-2026&utm_content=taissc\" target=\"_blank\" rel=\"noopener noreferrer\">Apply to the Searchlight &rarr;\x3C/a>\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\n\x3Chr>\n\x3Cp>The single thing to do this week: pick the one agent you run with the most access and write down everything it can touch. Honest worst case: you spend fifteen minutes confirming it is fine. Likely case: you find a credential it inherited months ago and has never used.\x3C/p>\n\x3Cp>Every play this newsletter has shipped, including the permission audit above once we turn it into a checklist on Saturday, lives with the people testing them in the free AI Academy.\x3C/p>\n\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" align=\"center\" style=\"margin:30px auto 10px auto;\">\x3Ctr>\x3Ctd class=\"kit-cta\" align=\"center\" bgcolor=\"#0074CA\" style=\"border-radius:6px;\">\x3Ca href=\"https://www.skool.com/try-ai-academy/about\" target=\"_blank\" rel=\"noopener noreferrer\">Join the free AI Academy &rarr;\x3C/a>\x3C/td>\x3C/tr>\x3C/table>\n\x3Cp style=\"text-align:center; margin:0 0 30px 0; font-size:13px; color:#5F5E5A;\">Free, no card. The tools and people behind this issue live here.\x3C/p>\n\x3Chr>\n\x3Cp>\x3Cstrong>That's the Debrief.\x3C/strong>\x3C/p>\n\x3Cp>See you Saturday for the Toolkit.\x3C/p>\n\x3Cp>From Drew and The AI Debrief team\x3C/p>\n\x3Chr>\n\x3Cp>\x3Cstrong>P.S.\x3C/strong> Which agent in your setup has the most access it doesn't need? Hit reply and tell us. We read every one.\x3C/p>\n\x3Cp>\x3Cstrong>P.P.S.\x3C/strong> New here, or skipped one? Every past edition is in the \x3Ca href=\"https://thedebrief.ai/archive\" target=\"_blank\" rel=\"noopener noreferrer\">archive\x3C/a>.\x3C/p>\x3C/div>\x3C/td>\x3C/tr>\x3C/tbody>\x3C/table>",edition:"debrief",issueNo:25},$R[28]={slug:"gemini-stopped-watching-the-whole-video",title:"Gemini stopped watching the whole video",subtitle:"Google's new video mode reads the transcript first and only pulls the frames it needs. Here's the notes machine to build on top of it.",date:"2026-09-04",thumbnail:"https://embed.filekitcdn.com/e/jeCKAoLJGmzsPEwMDzFkXA/g7zYL5bFYMBtvM1j4pzgQj",blocks:$R[29]=[],html:"\x3Ctable cellPadding=\"0\" cellSpacing=\"0\" style=\"width:100%;margin:0 auto\">\x3Ctbody>\x3Ctr>\x3Ctd>\x3Cdiv>\x3Cdiv>\x3Cp>Welcome to the Toolkit, the Friday half of The AI Debrief. Tuesday is what happened. Friday is what to do about it: one tool worth your time, one play you can run today, and nothing you have to buy.\x3C/p>\x3Cp>Here's what we're getting into this week. Tuesday was about who controls the model menu. Today is smaller and more useful: Google quietly changed how Gemini watches video, and it turns the most expensive kind of input into the cheapest way to take notes.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 8px 0;\">\x3Ctr>\x3Ctd style=\"border-top:1px solid #E5E5E5; border-bottom:1px solid #E5E5E5; padding:14px 2px;\">\x3Cp style=\"margin:0 0 8px 0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#5F5E5A;\">In this email\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; The video mode that reads the transcript first and only loads the frames it needs\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; A drop-a-link, get-timestamped-notes pipeline you can build before lunch\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; What it costs, which numbers are Google's, and where it will let you down\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#127916; The tool: a model that skips the boring parts\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"margin:16px 0 8px 0; font-size:19px; font-weight:700; line-height:1.3; color:#0A0A0A;\">Gemini agentic video understanding\x3C/p>\x3Cp>Every video you have ever handed to a model was processed the same way: sampled into frames at a fixed rate, every frame turned into tokens, whether or not anything happened in it. A one-hour screen recording of someone talking over a static slide cost the same as an hour of action. That is 
1why \"just summarize this video\" has always been a bill you do not run twice.\x3C/p>\x3Cp>On September 1 Google switched on agentic video understanding across Gemini 3.8 Flash, 3.7 Flash, 3.6 Flash and 3.5 Flash-Lite. You flip one setting, \x3Ccode>processing: agentic\x3C/code> on the video input, and the model stops sampling and starts navigating. It reads the transcript first, decides which stretches matter for your prompt, and loads only those frames or audio segments, adjusting frame rate and resolution as it goes. Ask for \"every moment the speaker shows a pricing table\" and it does not watch the hour. It finds the tables.\x3C/p>\x3Ctable class=\"kit-stack\" role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:22px 0 22px 0;\">\x3Ctr>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#F6F6F4; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#5F5E5A;\">Tokens, long video\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0A0A0A;\">up to 88% fewer\x3C/p>\x3C/td>\x3Ctd width=\"2%\">&nbsp;\x3C/td>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#E6F1FB; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#0074CA;\">Max length\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0074CA;\">3 hours\x3C/p>\x3C/td>\x3Ctd width=\"2%\">&nbsp;\x3C/td>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#F6F6F4; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#5F5E5A;\">YouTube, free tier\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0A0A0A;\">8 hrs/day\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Cp>Google's numbers: up to 88% fewer total tokens and roughly 7% higher quality on long-form content versus the old static mode. Those are Google's numbers on Google-chosen videos, so hold them the way we held Cloudflare's last week. The mechanism is the part I believe, because it is the same trick that made every good agent cheaper this year: do not read everything, read the index, then read what matters.\x3C/p>\x3Cp>The parts that are not marketing. The navigation happens as thought tokens and whatever it loads is billed as tool-use tokens, so the bill floats with how hard your question is, and a vague prompt makes it wander. It is in Preview and Google says costs are likely to change. Limits: the 1M-context models take videos up to three hours at low resolution or one hour at high; files go in through the File API at 2 GB each on the free tier and 20 GB paid; YouTube URLs go in directly, capped at eight hours of video a day on the free tier, uncapped on paid, up to ten links in a single request.\x3C/p>\x3Cp>\x3Cstrong>Who this is not for.\x3C/strong> Anything where every frame matters: QA of a UI, counting things, anything forensic. Static mode still exists for that and you should keep using it. This is for one question, \"what did they say and when\", asked of content too long to watch.\x3C/p>\x3Cp>\x3Ca href=\"https://ai.google.dev/gemini-api/docs/video-understanding\" target=\"_blank\" rel=\"noopener noreferrer\">\x3Cstrong>Video understanding docs &rarr;\x3C/strong>\x3C/a> &nbsp;&middot;&nbsp; \x3Ca href=\"https://ai.google.dev/gemini-api/docs/changelog\" target=\"_blank\" rel=\"noopener noreferrer\">the Sep 1 changelog entry\x3C/a>\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#9889; Steal this: a video-to-notes machine, ~20 minutes\x3C/td>\x3C/tr>\x3C/table>\x3Cp>What it replaces: a transcription subscription, or the forty minutes of scrubbing per video you do instead. What it costs after: nothing, as long as you stay inside eight hours of YouTube a day on the free tier.\x3C/p>\x3Cp>\x3Cstrong>1. Get a key.\x3C/strong> \x3Ca href=\"https://aistudio.google.com/\" target=\"_blank\" rel=\"noopener noreferrer\">Google AI Studio\x3C/a>, Get API key. The free tier is enough for everything below.\x3C/p>\x3Cp>\x3Cstrong>2. Write the prompt once, and make it demand timestamps.\x3C/strong> This is the whole play:\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:16px 0;\">\x3Ctr>\x3Ctd class=\"kit-code\" style=\"background-color:#0A0A0A; padding:16px 18px;\">\x3Cp style=\"margin:0; font-family:Menlo,Consolas,monospace; font-size:12px; line-height:1.7; color:#F6F6F4; word-break:break-word;\">You are taking notes on this video for someone who will not watch it. Return, in this order: (1) a three-sentence summary; (2) every concrete claim, number, price, or tool name, each with its timestamp; (3) any step-by-step instructions shown or spoken, as a numbered list with timestamps; (4) the three moments worth watching in full, with timestamps and one line on why. Do not paraphrase numbers. If there are no instructions, say so.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Cp>\x3Cstrong>3. One call.\x3C/strong> Model \x3Ccode>gemini-3.8-flash\x3C/code>, the YouTube URL as the video input with \x3Ccode>processing\x3C/code> set to \x3Ccode>agentic\x3C/code>, the prompt as the text part. The docs page above shows the exact shape for Python, JavaScript and plain REST; it is about twelve lines in any of them.\x3C/p>\x3Cp>\x3Cstrong>4. Save the reply as a Markdown file named after the video\x3C/strong>, into whichever folder your notes app already watches. Obsidian, Notion, a plain directory, it does not matter. The point is that the notes land where you will actually see them.\x3C/p>\x3Cp>\x3Cstrong>5. Batch it.\x3C/strong> Put your links in a list and loop, one call per video with a short pause between them. Up to ten links can share one request, which is how you ask cross-video questions: \"which of these five tutorials actually shows the setup step?\"\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 18px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #0A0A0A; padding:14px 16px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0A0A0A;\">Why this one works\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:15px; line-height:1.6; color:#0A0A0A;\">Timestamps. A summary you cannot verify is a summary you will not trust, and after two bad ones you stop reading them. Timestamps make every line checkable in ten seconds and turn the notes into an index of the video instead of a replacement for it. Asking for \"the three moments worth watching in full\" is the same idea: the model earns your trust by pointing at the video, not by standing in for it.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Cp>\x3Cstrong>Why it fits the free tier:\x3C/strong> eight hours of YouTube a day is more than you watch. Static mode would have burned through that allowance on the first long video. Agentic mode skips, which is the whole reason this is a free play now and was not a month ago.\x3C/p>\x3Cp>\x3Cstrong>What it saves:\x3C/strong> call it thirty minutes for every hour-long video you would otherwise scrub, plus the ones you were never going to get to.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#128250; Go deeper\x3C/td>\x3C/tr>\x3C/table>\x3Cp>The full script, the prompt, and a second version that watches a YouTube channel's feed and files notes on every new upload automatically are going in the vault.\x3C/p>\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" align=\"center\" style=\"margin:30px auto 10px auto;\">\x3Ctr>\x3Ctd class=\"kit-cta\" align=\"center\" bgcolor=\"#0074CA\" style=\"border-radius:6px;\">\x3Ca href=\"https://www.skool.com/try-ai-academy/classroom/8cd10fa9?md=3b6dac02472c41f998a8da8fc509af40\" target=\"_blank\" rel=\"noopener noreferrer\">Get the build &rarr;\x3C/a>\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"text-align:center; margin:0 0 30px 0; font-size:13px; color:#5F5E5A;\">Free. Same place the rest of the Toolkit builds live.\x3C/p>\x3Cp>If you build only one piece of this, run the single call on a video you have already watched. Compare the notes to what you remember. That tells you in five minutes whether to trust it with the ones you have not. Honest worst case: the preview pricing changes, and you own a 
1script that costs a few cents more per video than it did this week.\x3C/p>\x3Chr>\x3Cp>\x3Cstrong>That's the Toolkit.\x3C/strong>\x3C/p>\x3Cp>See you Tuesday for the Debrief.\x3C/p>\x3Cp>From Drew and The AI Debrief team\x3C/p>\x3Chr>\x3Cp>\x3Cstrong>P.S.\x3C/strong> What's the video you keep meaning to watch and never do - a two-hour tutorial, a conference talk, an earnings call? Hit reply and tell us. We read every one.\x3C/p>\x3Cp>\x3Cstrong>P.P.S.\x3C/strong> New here, or skipped one? Every past edition is in the \x3Ca href=\"https://thedebrief.ai/archive\" target=\"_blank\" rel=\"noopener noreferrer\">archive\x3C/a>.\x3C/p>\x3C/div>\x3C/div>\x3C/td>\x3C/tr>\x3C/tbody>\x3C/table>",edition:"toolkit",issueNo:24},$R[30]={slug:"cursor-just-lost-openai",title:"Cursor just lost OpenAI",subtitle:"OpenAI gave Cursor 76 days' notice. The lesson applies to every tool that rents its brain from somebody else.",date:"2026-09-01",thumbnail:"https://embed.filekitcdn.com/e/jeCKAoLJGmzsPEwMDzFkXA/gpjWPTfJDJYD2Qcmyx9iAQ",blocks:$R[31]=[],html:"\x3Ctable cellPadding=\"0\" cellSpacing=\"0\" style=\"width:100%;margin:0 auto\">\x3Ctbody>\x3Ctr>\x3Ctd>\x3Cdiv>\x3Cp>Welcome to the Debrief, the Tuesday half of The AI Debrief. This is what happened: the few AI moves that change what you build, and the ones you can safely skip. Friday is the Toolkit, where we hand you something to use.\x3C/p>\x3Cp>Here's what we're getting into this week. One of the most used AI coding tools on earth just learned its biggest model supplier is leaving, and a courtroom fight we have covered since February finally produced a ruling.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 8px 0;\">\x3Ctr>\x3Ctd style=\"border-top:1px solid #E5E5E5; border-bottom:1px solid #E5E5E5; padding:14px 2px;\">\x3Cp style=\"margin:0 0 8px 0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#5F5E5A;\">In this email\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; The November 12 shutoff, and the dependency audit it should trigger this week\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; The Pentagon ruling that answers the question we left open in May\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; A 1M-context model you can now download under MIT\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#128268; The model menu is a supply chain\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"margin:16px 0 8px 0; font-size:19px; font-weight:700; line-height:1.3; color:#0A0A0A;\">OpenAI is cutting Cursor off on November 12\x3C/p>\x3Cp>On August 28 \x3Ca href=\"https://openai.com/index/our-decision-on-cursor-following-its-acquisition-by-spacex/\" target=\"_blank\" rel=\"noopener noreferrer\">OpenAI announced\x3C/a> it is winding down the contract that puts its models inside Cursor, following Cursor's acquisition by SpaceX. Direct access ends November 12, the longest notice the contract allows. Future models, including Astra, will not be offered to Cursor at all. The stated reason is blunt: after X breached OpenAI's terms and xAI admitted to violating them, OpenAI says it \"cannot be confident that SpaceX will use our technology within our terms of service.\"\x3C/p>\x3Ctable class=\"kit-stack\" role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:22px 0 22px 0;\">\x3Ctr>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#F6F6F4; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#5F5E5A;\">Notice given\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0A0A0A;\">Aug 28\x3C/p>\x3C/td>\x3Ctd width=\"2%\">&nbsp;\x3C/td>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#E6F1FB; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#0074CA;\">Access ends\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0074CA;\">Nov 12\x3C/p>\x3C/td>\x3Ctd width=\"2%\">&nbsp;\x3C/td>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#F6F6F4; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#5F5E5A;\">Partnership\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0A0A0A;\">~4 years\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Cp>What actually changes: GPT models leave Cursor's model picker after November 12, ending a partnership that ran nearly four years. Claude, Grok and the rest stay. Nothing breaks today, and OpenAI says it will support affected developers through the transition, so the 76 days are a real runway rather than a fire drill.\x3C/p>\x3Cp>The Astra line is the detail that ages. A shutoff of today's models is a one-time migration; being excluded from future models means the gap between Cursor-with-OpenAI and Cursor-without widens with every release from here. OpenAI called out nearly four years of partnership and \"enormous respect\" for the team in the same post, which tells you this was a legal-risk decision, not a product one. The trigger was a trust judgment: two of Elon Musk's companies had already broken OpenAI's terms, and OpenAI decided a third chance was not worth underwriting.\x3C/p>\x3Cp>The direction of travel was already visible two days earlier. On August 26, SpaceXAI pushed Grok Bot onto Cursor Pro and Cursor Teams with its own separate usage quota. Put the two moves side by side and the shape is clear: Cursor is becoming a Grok surface, and the supplier it was built on for four years is leaving before that finishes happening.\x3C/p>\x3Cp>The feud is not the story, and I have read a week of takes that treat it as one. The story is that a dispute between two vendors reached into a third company's product and removed capability from millions of paying developers, on a timeline set by a contract clause. Every AI tool you pay for is a reseller of somebody else's intelligence, and reseller agreements have politics in them. It will not be the last time one of them ends this way.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 18px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #0A0A0A; padding:14px 16px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0A0A0A;\">Why you care\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:15px; line-height:1.6; color:#0A0A0A;\">If GPT models are load-bearing anywhere in your Cursor setup, you have until November 12, and switching costs are lowest while nothing is on fire. Run one real task against a second model family this week and note what breaks. The
1n ask the same question of every tool you rely on that rents its models, which is nearly all of them: what happens to your workflow if one supplier walks?\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#9889; Worth your time\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"margin-top:16px;\">\x3Cstrong>The Pentagon blacklist lost in court.\x3C/strong>\x3Cbr>On August 28, U.S. District Judge Rita Lin \x3Ca href=\"https://www.cnbc.com/2026/08/28/judge-blocks-pentagon-blacklist--anthropic-.html\" target=\"_blank\" rel=\"noopener noreferrer\">struck down\x3C/a> the Pentagon's designation of Anthropic as a supply-chain risk. The 59-page ruling found the label was retaliation for Anthropic refusing to allow Claude to be used for mass surveillance of Americans and fully autonomous weapons: \"The empty invocation of national security is not a blank check to punish and retaliate against government critics.\" We have covered this fight since February and left it open in May with oral arguments pending; this is the answer. It is not the end - the Pentagon is expected to appeal, and a second designation is still in front of the D.C. Circuit - but there is now a merits ruling saying an agency cannot spend \"national security\" like a currency against a vendor that told it no. If anyone on your side ever asks whether a model vendor's stated red lines survive contact with its biggest customer, this is the citation.\x3C/p>\x3Cp>\x3Cstrong>\"Ox Alpha\" was Z.ai, and it's MIT.\x3C/strong>\x3Cbr>The anonymous model sitting on OpenRouter since August 20 turned out to be \x3Ca href=\"https://huggingface.co/zai-org/GLM-5.3-Flash\" target=\"_blank\" rel=\"noopener noreferrer\">GLM-5.3-Flash\x3C/a>: a natively multimodal 320B mixture-of-experts with 18B active parameters and a 1M-token context window, weights on Hugging Face under MIT. Z.ai's headline benchmark, a DeepSWE score of 63.4 against its predecessor's 46.2, is vendor-reported, so hold it as a claim. Two things are not claims: the license, and the price, $0.15 in and $0.50 out per million tokens, half that until September 9. One wrinkle to read before you build: the full-size GLM-5.3 weights that followed on August 27 carry a restrictive custom license. The Flash tier is the MIT one, and for an 18B-active model that is the tier you would self-host anyway.\x3C/p>\x3Cp>\x3Cstrong>Amazon bought the company behind DuckDB.\x3C/strong>\x3Cbr>AWS signed a definitive agreement to acquire DuckLabs on August 26. DuckDB itself \x3Ca href=\"https://duckdb.org/2026/08/26/ducklabs-to-join-aws\" target=\"_blank\" rel=\"noopener noreferrer\">stays MIT-licensed\x3C/a> under the independent DuckDB Foundation: Amazon bought the employer, not the project. Nothing in your data stack changes today; the thing to watch is where the paid features land.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#128581; Skip it\x3C/td>\x3C/tr>\x3C/table>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:16px 0 18px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #00BC7D; padding:16px 18px;\">\x3Cp style=\"margin:0 0 8px 0; font-size:17px; font-weight:700; color:#0A0A0A;\">Nvidia's record quarter, and the week of \"inference crown\" claims around it\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.6; color:#0A0A0A;\">Nvidia reported another record quarter on August 26, bracketed by a new inference accelerator and a round of dueling benchmark claims about who serves tokens fastest. We filed the last giant Nvidia number under Skip it and the position holds: capacity numbers this 
1size are macro news, not build decisions. Nothing in your stack gets faster, cheaper or safer because a quarter broke a record. The one number from earnings week worth sitting with points the other way - an NBER survey found 89% of executives report no measurable AI productivity gains. Gains follow workflow redesign, not headlines.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Chr>\x3Cp>The single thing to do this week: count how many of the models in whatever you build in come from one supplier, then run one real task on the second-place option. Honest worst case: you spend twenty minutes confirming your fallback works, and November 12 becomes somebody else's problem.\x3C/p>\x3Cp>Every play this newsletter has shipped, including Friday's competitor-watcher build, lives with the people testing them in the free AI Academy.\x3C/p>\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" align=\"center\" style=\"margin:30px auto 10px auto;\">\x3Ctr>\x3Ctd class=\"kit-cta\" align=\"center\" bgcolor=\"#0074CA\" style=\"border-radius:6px;\">\x3Ca href=\"https://www.skool.com/try-ai-academy/about\" target=\"_blank\" rel=\"noopener noreferrer\">Join the free AI Academy &rarr;\x3C/a>\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"text-align:center; margin:0 0 30px 0; font-size:13px; color:#5F5E5A;\">Free, no card. The tools and people behind this issue live here.\x3C/p>\x3Chr>\x3Cp>\x3Cstrong>That's the Debrief.\x3C/strong>\x3C/p>\x3Cp>See you Friday for the Toolkit.\x3C/p>\x3Cp>From Drew and The AI Debrief team\x3C/p>\x3Chr>\x3Cp>\x3Cstrong>P.S.\x3C/strong> What breaks in your setup if one model supplier walks? Hit reply and tell us. We read every one.\x3C/p>\x3Cp>\x3Cstrong>P.P.S.\x3C/strong> New here, or skipped one? Every past edition is in the \x3Ca href=\"https://thedebrief.ai/archive\" target=\"_blank\" rel=\"noopener noreferrer\">archive\x3C/a>.\x3C/p>\x3C/div>\x3C/td>\x3C/tr>\x3C/tbody>\x3C/table>",edition:"debrief",issueNo:23},$R[32]={slug:"the-16-tool-you-can-delete",title:"The $16 tool you can delete",subtitle:"Cloudflare built a browser with no Chrome inside it. Here's the competitor-watcher you can put on top of it in an afternoon.",date:"2026-08-28",thumbnail:"https://embed.filekitcdn.com/e/jeCKAoLJGmzsPEwMDzFkXA/s3C5EUDRNKfV8wworkeAcV",blocks:$R[33]=[],html:"\x3Cp>Welcome to the Toolkit, the Friday half of The AI Debrief. Tuesday tells you what happened; Friday hands you something to use. (If you missed it, Tuesday's issue was about four betas ending at once, and the one of them that changes your request shape.)\x3C/p>\x3Cp>Tuesday ended on a browser tool that reads page structure instead of pixels. That idea is the good one, and you do not need a frontier model bill to try it. This week's tool is the cheapest possible way to get your hands on it.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 8px 0;\">\x3Ctr>\x3Ctd style=\"border-top:1px solid #E5E5E5; border-bottom:1px solid #E5E5E5; padding:14px 2px;\">\x3Cp style=\"margin:0 0 8px 0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#5F5E5A;\">In this email\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; The browser that runs without Chromium, and what that buys\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; A daily competitor-pricing watcher you can build in about 40 minutes\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; The honest list of what this thing cannot do\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#129520; The tool: a browser with no browser in it\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"margin:16px 0 8px 0; font-size:19px; font-weight:700; line-height:1.3; color:#0A0A0A;\">Cloudflare Kitesurf\x3C/p>\x3Cp>Every headless-browser product you have used works roughly the same way: it starts a real Chromium process per session, and 
1you pay for that process in CPU, memory and money. Kitesurf, which Cloudflare launched on August 6, does not ship Chromium at all. It runs inside V8 isolates on Workers.\x3C/p>\x3Ctable class=\"kit-stack\" role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:22px 0 22px 0;\">\x3Ctr>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#F6F6F4; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#5F5E5A;\">CPU\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0A0A0A;\">3.8&times; less\x3C/p>\x3C/td>\x3Ctd width=\"2%\">&nbsp;\x3C/td>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#E6F1FB; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#0074CA;\">Memory\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0074CA;\">7&times; less\x3C/p>\x3C/td>\x3Ctd width=\"2%\">&nbsp;\x3C/td>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#F6F6F4; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#5F5E5A;\">Wall time\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0A0A0A;\">1.7&times; slower\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Cp>There are three pieces. The \x3Cstrong>Engine\x3C/strong> is the public-facing part. It speaks Chrome DevTools Protocol, so existing tooling can talk to it, and it holds session state. \x3Cstrong>PageScript\x3C/strong> spins up a Dynamic Worker per page to handle the DOM and run JavaScript and WASM. \x3Cstrong>PageRenderer\x3C/strong> does the actual rasterizing using Rust libraries compiled to WASM: Blitz for layout, Firefox's Stylo engine for CSS, Parley for text.\x3C/p>\x3Cp>That architecture is the whole story. No per-session Chromium process means the cost curve is different, and Cloudflare publishes the comparison: 3.1&times; less CPU on screenshots and 3.8&times; less on HTML extraction, 4.7&times; less memory on screenshots and 7.0&times; less on extraction. These are vendor-reported numbers on vendor-chosen workloads, so hold them loosely. But Cloudflare also publishes the tradeoff, which is the part that makes me believe the rest. Kitesurf is \x3Cstrong>1.7 to 1.8 times slower in wall time\x3C/strong>. It is cheaper, not faster. For anything on a cron that is a trade you will take every time; for anything a user is waiting on, it is not.\x3C/p>\x3Cp>It passes over 235,000 Web Platform Test subtests, which is a real number for a rendering engine written from scratch. There is a \x3Ca href=\"https://kitesurf.cloudflare.app/\" target=\"_blank\" rel=\"noopener noreferrer\">public playground\x3C/a> if you want to watch it render something before you write any code.\x3C/p>\x3Cp>Getting at it is a query parameter. Add \x3Ccode>?browser=kitesurf\x3C/code> to any \x3Ca href=\"https://developers.cloudflare.com/browser-run/quick-actions/\" target=\"_blank\" rel=\"noopener noreferrer\">Quick Action endpoint\x3C/a> and you are on it: \x3Ccode>/markdown\x3C/code>, \x3Ccode>/content\x3C/code>, \x3Ccode>/screenshot\x3C/code>, \x3Ccode>/pdf\x3C/code>, \x3Ccode>/snapshot\x3C/code>, \x3Ccode>/accessibilityTree\x3C/code>, \x3Ccode>/scrape\x3C/code>, \x3Ccode>/links\x3C/code>, \x3Ccode>/crawl\x3C/code>, and \x3Ccode>/json\x3C/code> for AI structured extraction. There is a CDP WebSocket endpoint and an MCP config too.\x3C/p>\x3Cp>\x3Cstrong>Cost.\x3C/strong> Free during beta, behind per-account limits Cloudflare has not published, so treat the free ride as real but unquantified. Underneath, \x3Ca href=\"https://developers.cloudflare.com/browser-run/pricing/\" target=\"_blank\" rel=\"noopener noreferrer\">Browser Run billing\x3C/a> applies: the Workers free tier gives 10 minutes of browser time a day and 3 concurrent browsers, with \x3Ccode>/crawl\x3C/code> capped at 5 jobs a day across 100 pages each. Workers Paid includes 10 browser-hours a month and then meters at \x3Cstrong>$0.09 an hour\x3C/strong>, with 10 concurrent browsers included and \x3Cstrong>$2.00\x3C/strong> per browser beyond that. Requests that fail on \x3Ccode>waitForTimeout\x3C/code> are not charged.\x3C/p>\x3Cp>\x3Cstrong>Who this is not for.\x3C/strong> No video. No WebGL. It cannot get through TLS-fingerprint bot challenges. It does not hold long-lived authenticated sessions. This is stateless, one-shot work: fetch a page, extract something, leave. If you need to log in and stay logged in, keep using what you have.\x3C/p>\x3Cp>\x3Ca href=\"https://developers.cloudflare.com/browser-run/kitesurf/\" target=\"_blank\" rel=\"noopener noreferrer\">\x3Cstrong>Kitesurf docs &rarr;\x3C/strong>\x3C/a>\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#9889; Steal this: a competitor-pricing watcher, ~40 minutes\x3C/td>\x3C/tr>\x3C/table>\x3Cp>What it replaces: a Firecrawl Hobby plan at $16 a month plus a Zapier monitor. What it costs after: nothing, or about $5 if you are already on Workers Paid.\x3C/p>\x3Cp>\x3Cstrong>1. Get a token.\x3C/strong> Cloudflare dashboard, API Tokens, create one with \x3Cstrong>Browser Rendering - Edit\x3C/strong>. Copy your Account ID while you are there.\x3C/p>\x3Cp>\x3Cstrong>2. Smoke-test before writing any code.\x3C/strong> One curl proves the whole chain works:\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:16px 0;\">\x3Ctr>\x3Ctd class=\"kit-code\" style=\"background-color:#0A0A0A; padding:16px 18px;\">\x3Cp style=\"margin:0; font-family:Menlo,Consolas,monospace; font-size:12px; line-height:1.7; color:#F6F6F4; word-break:break-all;\">&#99;url -X POST 'https&#58;//api&#46;cloudflare&#46;com/client/v4/accounts/&lt;ACCOUNT_ID&gt;/browser-run/markdown?browser=kitesurf' \\\x3Cbr>&nbsp;&nbsp;-H '&#65;uthorization: &#66;earer &lt;TOKEN&gt;' \\\x3Cbr>&nbsp;&nbsp;-H 'Content-Type: application/json' \\\x3Cbr>&nbsp;&nbsp;
1-d '{\"url\":\"https&#58;//competitor&#46;com/pricing\"}'\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Cp>\x3Cstrong>3. Scaffold it.\x3C/strong> \x3Ccode>npm create cloudflare@latest watcher\x3C/code>. Add a KV namespace binding called \x3Ccode>SEEN\x3C/code> and a Cron Trigger of \x3Ccode>0 13 * * *\x3C/code>, daily at 13:00 UTC.\x3C/p>\x3Cp>\x3Cstrong>4. The logic is about fifteen lines.\x3C/strong> For each URL in a hardcoded array: call \x3Ccode>/markdown\x3C/code> with \x3Ccode>browser=kitesurf\x3C/code>, \x3Ccode>await env.SEEN.get(url)\x3C/code>, compare against what came back, \x3Ccode>put()\x3C/code> the new version. If it changed, hand the old and new text to a cheap model with this prompt:\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:16px 0;\">\x3Ctr>\x3Ctd class=\"kit-code\" style=\"background-color:#0A0A0A; padding:16px 18px;\">\x3Cp style=\"margin:0; font-family:Menlo,Consolas,monospace; font-size:12px; line-height:1.7; color:#F6F6F4; word-break:break-all;\">You are a competitive-intel filter. Below are two versions of a page, BEFORE and AFTER. Report ONLY material business changes: price changes, new or removed plans, changed limits or quotas, new features, changed terms. Ignore copy edits, testimonials, blog links, dates, and formatting. If nothing material changed, reply exactly: NO CHANGE. Otherwise reply with up to 5 bullets, each naming the old value and the new value.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Cp>\x3Cstrong>5. If the reply is not NO CHANGE, POST it to a Slack webhook.\x3C/strong> Attach a \x3Ccode>/screenshot?browser=kitesurf\x3C/code> if you want the visual.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 18px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #0A0A0A; padding:14px 16px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0A0A0A;\">Why this one works\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:15px; line-height:1.6; color:#0A0A0A;\">The filter prompt is the part that makes this worth building rather than abandoning. A raw diff on a marketing page fires every time somebody swaps a testimonial, and you will mute the channel inside a week. Forcing a model to answer “material or not” and to name the old value and the new value is what turns it from noise into something you actually read.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Cp>\x3Cstrong>Why it fits the free tier:\x3C/strong> twenty pages at roughly two seconds each is about forty seconds a day against a ten-minute daily allowance. You only get 3 concurrent browsers, so loop sequentially or chunk in threes. Do not fire twenty at once and wonder why it fails.\x3C/p>\x3Cp>\x3Cstrong>What it saves:\x3C/strong> a manual competitor sweep you either do badly or skip entirely. Call it 45 minutes a week, plus finding out about a price change on the day rather than the quarter.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#128250; Go deeper\x3C/td>\x3C/tr>\x3C/table>\x3Cp>The full worker file, the KV setup and the Slack payload are going in the vault, along with a second version that watches changelogs instead of pricing pages.\x3C/p>\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" align=\"center\" style=\"margin:30px auto 10px auto;\">\x3Ctr>\x3Ctd class=\"kit-cta\" align=\"center\" bgcolor=\"#0074CA\" style=\"border-radius:6px;\">\x3Ca href=\"https://www.skool.com/try-ai-academy/classroom/8cd10fa9?md=3b6dac02472c41f998a8da8fc509af40\" target=\"_blank\" rel=\"noopener noreferrer\">Get the build &rarr;\x3C/a>\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"text-align:center; margin:0 0 30px 0; font-size:13px; color:#5F5E5A;\">Free. Same place the rest of the Toolkit builds live.\x3C/p>\x3Cp>If you build only one piece of this, build step 2, the single curl. It takes ninety seconds and it tells you whether this whole category is useful to you before you commit an afternoon. Honest worst case: Kitesurf's undisclosed beta limits turn out to be tighter than your use, and you have lost ninety seconds finding out.\x3C/p>\x3Chr>\x3Cp>\x3Cstrong>That's the Toolkit.\x3C/strong>\x3C/p>\x3Cp>See you Tuesday for the Debrief.\x3C/p>\x3Cp>From Drew and The AI Debrief team\x3C/p>\x3Chr>\x3Cp>\x3Cstrong>P.S.\x3C/strong> What page would you actually want watched: a competitor's pricing, a supplier's terms, a job board? Hit reply and tell us. We read every one.\x3C/p>\x3Cp>\x3Cstrong>P.P.S.\x3C/strong> New here, or skipped one? Every past edition is in the \x3Ca href=\"https://thedebrief.ai/archive\" target=\"_blank\" rel=\"noopener noreferrer\">archive\x3C/a>.\x3C/p>",edition:"toolkit",issueNo:22},$R[34]={slug:"three-of-these-are-safe-to-ignore-25561566",title:"Three of these are safe to ignore",subt
1itle:"Computer use, Agent Skills, the Files API and the Python SDK all changed across two days. Three of them are backward compatible. One is not.",date:"2026-08-25",thumbnail:"",blocks:$R[35]=[$R[36]={c:"The Debrief &middot; No. 21 &middot; Tue Aug 25",t:"p"},$R[37]={c:"Three of these are safe to ignore",t:"p"},$R[38]={c:"Computer use, Agent Skills, the Files API and the Python SDK all changed across two days. Three of them are backward compatible. One is not.",t:"p"},$R[39]={c:"\x3Cstrong>Quick note before we start:\x3C/strong> you subscribed to The AI Debrief a while back, and then things went quiet on you. That's on me - we moved the whole newsletter to a new platform, and I didn't want to hit your inbox until it was worth it. This is the new era: the Debrief every Tuesday, the Toolkit every Friday. And if AI emails aren't your thing anymore, no hard feelings - the unsubscribe link at the bottom works on the first click.",t:"p"},$R[40]={c:"Welcome to the Debrief, the Tuesday half of The AI Debrief. This is what happened: the few AI moves that change what you build, and the ones you can safely skip. Friday is the Toolkit, where we hand you something to use.",t:"p"},$R[41]={c:"Here's what we're getting into this week. A stack of things you may already depend on stopped being experimental on August 19 and 20, and exactly one of them will not run on your existing request shape.",t:"p"},$R[42]={c:"In this email",t:"p"},$R[43]={c:"&raquo; The migration hiding inside a GA announcement",t:"p"},$R[44]={c:"&raquo; A browser tool that reads page structure instead of pixels",t:"p"},$R[45]={c:"&raquo; Why data retention is turning into a procurement question",t:"p"},$R[46]={c:"Anthropic's agent stack went GA on two days, not one",t:"p"},$R[47]={c:"On August 19 the Files API and Agent Skills left beta. On August 20 \x3Ca href=\"https://claude.com/blog/computer-use-skills-api-files-api\" target=\"_blank\" rel=\"noopener noreferrer\">computer use followed\x3C/a>, along with a new browser tool and a major version of the Python SDK. Most of the coverage treated it as one launch. It reads better as four separate migrations with four separate risk levels.",t:"p"},$R[48]={c:"Media per request",t:"p"},$R[49]={c:"100 &rarr; 600",t:"p"},$R[50]={c:"Actions per turn",t:"p"},$R[51]={c:"1 &rarr; many",t:"p"},$R[52]={c:"Python",t:"p"},$R[53]={c:"3.9 &rarr; 3.10+",t:"p"},$R[54]={c:"Computer use is now \x3Cstrong>computer_toolset_20260801\x3C/strong>, and it replaces the beta \x3Cstrong>computer_20251124\x3C/strong>. This one is not a header swap. Anthropic's own release notes say upgrading “changes the request shape and tool handling,” and they publish a dedicated \x3Ca href=\"https://platform.claude.com/docs/en/agents-and-tools/tool-use/computer-use-tool#migrate-from-computer-20251124\" target=\"_blank\" rel=\"noopener noreferrer\">migration guide\x3C/a> for it. What you get in exchange is the thing that actually matters for cost: batch actions, so the model can take several actions per turn instead of one per model call. zoom is on by default now, and there is per-member configuration through configs.",t:"p"},$R[55]={c:"The other three are gentler. The Files API dropped the files-api-2025-04-14 header, added file expiry through expires_in_seconds and expires_at, added pagination, and raised the media ceiling from 100 to 600 images or PDF pages per request on the 1M-token context window. Agent Skills dropped skills-2025-10-02 and picked up a /v1/skills endpoint that loads through the container parameter. Both keep working unchanged if you still send the old header, which is the detail worth holding onto.",t:"p"},$R[56]={c:"The \x3Ca href=\"https://github.com/anthropics/anthropic-sdk-python/blob/main/MIGRATION.md\" target=\"_blank\" rel=\"noopener noreferrer\">Python SDK\x3C/a> is the one that will surprise people, because nothing announces it. v1.0 moved from httpx to httpx2, requires Python 3.10 or later, removed the legacy Text Completions API, and removed temperature, top_p and top_k from Messages methods. Async .with_raw_response results now need await response.parse(). And AnthropicBedrock raise
1s an error when no AWS region is configured instead of quietly defaulting to us-east-1, which is a better behaviour and also a behaviour change in something that used to work.",t:"p"},$R[57]={c:"All of it runs on Fable 5, Mythos 5, Opus 5, Sonnet 5 and Opus 4.8.",t:"p"},$R[58]={c:"Why you care",t:"p"},$R[59]={c:"GA announcements read like good news, and this one mostly is. But the risk here is not evenly distributed, and the announcement does not sort it for you. Three of these four changes are backward compatible; one changes the shape of your request. Grep for \x3Cstrong>computer_20251124\x3C/strong> and treat anything you find as scheduled work. The two header drops can wait for a quiet afternoon. The SDK will announce itself the next time you build, and it will do it at the worst possible moment.",t:"p"},$R[60]={c:"\x3Ca href=\"https://platform.claude.com/docs/en/release-notes/api\" target=\"_blank\" rel=\"noopener noreferrer\">\x3Cstrong>Read the release notes &rarr;\x3C/strong>\x3C/a>",t:"p"},$R[61]={c:"\x3Cstrong>A browser tool that reads the page instead of looking at it.\x3C/strong> \x3Cstrong>browser_toolset_20260801\x3C/strong> is new, and it is the most interesting thing in the whole release. It drives a hosted browser by reading the accessibility tree - elements, forms, tabs - rather than hunting for pixels. It adds element references, form input, tab management and download reporting. The difference is not cosmetic. Pixel-targeted automation breaks when a button moves four pixels or a font renders differently; structure-targeted automation does not. That is roughly the line between a demo that works in a video and something you can leave running unattended. Friday's Toolkit has a way to try this idea for nothing.",t:"p"},$R[62]={c:"\x3Cstrong>Data retention quietly turned into a contract question.\x3C/strong> OpenAI published \x3Ca href=\"https://openai.com/index/offering-zero-data-retention-for-frontier-models/\" target=\"_blank\" rel=\"noopener noreferrer\">Private Safety Processing\x3C/a> on August 19. The mechanism: your content stays either on infrastructure you control, or on OpenAI servers encrypted with keys you hold, and OpenAI states its own personnel do not have a copy. When something is flagged, what comes back is “a narrowly defined signal indicating the type of activity involved,” not the prompt and not the response. Named design partners are Glean, Databricks, Abridge and Microsoft. Full rollout is September, with a technical white paper. Pricing is not disclosed. One caveat, because a lot of coverage is running it the other way: several outlets frame this as a direct shot at a competitor's retention policy, but OpenAI's own page says nothing about competitors at all. Treat the rivalry angle as reported, not stated. What is solid is the direction. If you sell into health, legal or finance, “which model” is drifting from a benchmark question toward a procurement one.",t:"p"},$R[63]={c:"\x3Cstrong>A DeepSeek update, not a new story.\x3C/strong> We covered the V4-Pro pricing restructure on August 18, so this is one line rather than a section: since August 23, all day Saturday and Sunday count as off-peak. If you already moved batch work to dodge the peak windows, you can now park the heaviest weekly job on a weekend at half rate. That is the entire change.",t:"p"},$R[64]={c:"\x3Cstrong>Unitree's IPO, and the “the robots are here” week it produced.\x3C/strong> The stock debuted on Shanghai's STAR Market on August 18, popped 629% intraday and closed up 487%, landing near $53B. The underlying business earned roughly &yen;278M on &yen;1.7B of 2025 revenue and shipped about 5,500 humanoids. That is somewhere near 1,300 times earnings and 210 times sales. It was priced at about $9B on August 6, which means the entire re-rating happened in retail order books over twelve days, on an offer oversubscribed thousands of times over. Nothing about what the robots can do changed in that window. It is a genuinely interesting market story and there is nothing in it you can build on.",t:"p"},$R[65]={c:"The one thing worth doing this week takes about thirty seconds: search your codebase for \x3Cstrong>computer_20251124\x3C/strong>. That is the only item here whose request shape has already changed. Everything else is a header you can drop when convenient, or a ver
1sion pin that will introduce itself later. Honest worst case if you ignore all of it - an agent that used to work starts throwing shape errors on a Tuesday, and you spend an hour finding out why.",t:"p"},$R[66]={c:"The 30-day challenge we opened on Friday is on Day 5, and it is built to be caught up on.",t:"p"},$R[67]={c:"Free, no card. Day 1 takes 20 minutes.",t:"p"},$R[68]={c:"\x3Cstrong>That's the Debrief.\x3C/strong>",t:"p"},$R[69]={c:"See you Friday for the Toolkit.",t:"p"},$R[70]={c:"From Drew and The AI Debrief team",t:"p"},$R[71]={c:"\x3Cstrong>P.S.\x3C/strong> What's still pinned to a beta header inside something you own? Hit reply and tell us. We read every one.",t:"p"},$R[72]={c:"\x3Cstrong>P.P.S.\x3C/strong> New here, or skipped one? Every past edition is in the \x3Ca href=\"https://thedebrief.ai/archive\" target=\"_blank\" rel=\"noopener noreferrer\">archive\x3C/a>.",t:"p"}],edition:"debrief",issueNo:null},$R[73]={slug:"three-of-these-are-safe-to-ignore",title:"Three of these are safe to ignore",subtitle:"Computer use, Agent Skills, the Files API and the Python SDK all changed across two days. Three of them are backward compatible. One is not.",date:"2026-08-25",thumbnail:"https://embed.filekitcdn.com/e/jeCKAoLJGmzsPEwMDzFkXA/8Cf4tgBsJRGSwUPdZ1GaeS",blocks:$R[74]=[],html:"\x3Cp>Welcome to the Debrief, the Tuesday half of The AI Debrief. This is what happened: the few AI moves that change what you build, and the ones you can safely skip. Friday is the Toolkit, where we hand you something to use.\x3C/p>\x3Cp>Here's what we're getting into this week. A stack of things you may already depend on stopped being experimental on August 19 and 20, and exactly one of them will not run on your existing request shape.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 8px 0;\">\x3Ctr>\x3Ctd style=\"border-top:1px solid #E5E5E5; border-bottom:1px solid #E5E5E5; padding:14px 2px;\">\x3Cp style=\"margin:0 0 8px 0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#5F5E5A;\">In this email\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; The migration hiding inside a GA announcement\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; A browser tool that reads page structure instead of pixels\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; Why data retention is turning into a procurement question\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#129521; The beta ended, and it took your request shape with it\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"margin:16px 0 8px 0; font-size:19px; font-weight:700; line-height:1.3; color:#0A0A0A;\">Anthropic's agent stack went GA on two days, not one\x3C/p>\x3Cp>On August 19 the Files API and Agent Skills left beta. On August 20 \x3Ca href=\"https://claude.com/blog/computer-use-skills-api-files-api\" target=\"_blank\" rel=\"noopener noreferrer\">computer use followed\x3C/a>, along with a new browser tool and a major version of the Python SDK. Most of the coverage treated it as one launch. It reads better as four separate migrations with four separate risk levels.\x3C/p>\x3Ctable class=\"kit-stack\" role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:22px 0 22px 0;\">\x3Ctr>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#F6F6F4; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#5F5E5A;\">Media per request\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0A0A0A;\">100 &rarr; 600\x3C/p>\x3C/td>\x3Ctd width=\"2%\">&nbsp;\x3C/td>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#E6F1FB; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#0074CA;\">Actions per turn\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0074CA;\">1 &rarr; many\x3C/p>\x3C/td>\x3Ctd width=\"2%\">&nbsp;\x3C/td>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#F6F6F4; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#5F5E5A;\">Python\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0A0A0A;\">3.9 &rarr; 3.10+\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Cp>Computer use is now \x3Cstrong>computer_toolset_20260801\x3C/strong>, and it replaces the beta \x3Cstrong>computer_20251124\x3C/strong>. This one is not a header swap. Anthropic's own release notes say upgrading “changes the request shape and tool handling,” and they publish a dedicated \x3Ca href=\"https://platform.claude.com/docs/en/agents-and-tools/tool-use/computer-use-tool#migrate-from-computer-20251124\" target=\"_blank\" rel=\"noopener noreferrer\">
1migration guide\x3C/a> for it. What you get in exchange is the thing that actually matters for cost: batch actions, so the model can take several actions per turn instead of one per model call. \x3Ccode>zoom\x3C/code> is on by default now, and there is per-member configuration through \x3Ccode>configs\x3C/code>.\x3C/p>\x3Cp>The other three are gentler. The Files API dropped the \x3Ccode>files-api-2025-04-14\x3C/code> header, added file expiry through \x3Ccode>expires_in_seconds\x3C/code> and \x3Ccode>expires_at\x3C/code>, added pagination, and raised the media ceiling from 100 to 600 images or PDF pages per request on the 1M-token context window. Agent Skills dropped \x3Ccode>skills-2025-10-02\x3C/code> and picked up a \x3Ccode>/v1/skills\x3C/code> endpoint that loads through the \x3Ccode>container\x3C/code> parameter. Both keep working unchanged if you still send the old header, which is the detail worth holding onto.\x3C/p>\x3Cp>The \x3Ca href=\"https://github.com/anthropics/anthropic-sdk-python/blob/main/MIGRATION.md\" target=\"_blank\" rel=\"noopener noreferrer\">Python SDK\x3C/a> is the one that will surprise people, because nothing announces it. v1.0 moved from \x3Ccode>httpx\x3C/code> to \x3Ccode>httpx2\x3C/code>, requires Python 3.10 or later, removed the legacy Text Completions API, and removed \x3Ccode>temperature\x3C/code>, \x3Ccode>top_p\x3C/code> and \x3Ccode>top_k\x3C/code> from Messages methods. Async \x3Ccode>.with_raw_response\x3C/code> results now need \x3Ccode>await response.parse()\x3C/code>. And \x3Ccode>AnthropicBedrock\x3C/code> raises an error when no AWS region is configured instead of quietly defaulting to \x3Ccode>us-east-1\x3C/code>, which is a better behaviour and also a behaviour change in something that used to work.\x3C/p>\x3Cp>All of it runs on Fable 5, Mythos 5, Opus 5, Sonnet 5 and Opus 4.8.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 18px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #0A0A0A; padding:14px 16px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0A0A0A;\">Why you care\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:15px; line-height:1.6; color:#0A0A0A;\">GA announcements read like good news, and this one mostly is. But the risk here is not evenly distributed, and the announcement does not sort it for you. Three of these four changes are backward compatible; one changes the shape of your request. Grep for \x3Cstrong>computer_20251124\x3C/strong> and treat anything you find as scheduled work. The two header drops can wait for a quiet afternoon. The SDK will announce itself the next time you build, and it will do it at the worst possible moment.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Cp>\x3Ca href=\"https://platform.claude.com/docs/en/release-notes/api\" target=\"_blank\" rel=\"noopener noreferrer\">\x3Cstrong>Read the release notes &rarr;\x3C/strong>\x3C/a>\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#9889; Worth your time\x3C/td>\x3C/tr>\x3C/table>\x3Cp>\x3Cstrong>A browser tool that reads the page instead of looking at it.\x3C/strong>\x3Cbr>\x3Cstrong>browser_toolset_20260801\x3C/strong> is new, and it is the most interesting thing in the whole release. It drives a hosted browser by reading the accessibility tree - elements, forms, tabs - rather than hunting for pixels. It adds element references, form input, tab management and download reporting. The difference is not cosmetic. Pixel-targeted automation breaks when a button moves four pixels or a font renders differently; structure-targeted automation does not. That is roughly the line between a demo that works in a video and something you can leave running unattended. Friday's Toolkit has a way to try this idea for nothing.\x3C/p>\x3Cp>\x3Cstrong>Data retention quietly turned into a contract question.\x3C/strong>\x3Cbr>OpenAI published \x3Ca href=\"https://openai.com/index/offering-zero-data-retention-for-frontier-models/\" target=\"_blank\" rel=\"noopener noreferrer\">Private Safety Processing\x3C/a> on August 19. The mechanism: your content stays either on infrastru
1cture you control, or on OpenAI servers encrypted with keys you hold, and OpenAI states its own personnel do not have a copy. When something is flagged, what comes back is “a narrowly defined signal indicating the type of activity involved,” not the prompt and not the response. Named design partners are Glean, Databricks, Abridge and Microsoft. Full rollout is September, with a technical white paper. Pricing is not disclosed. One caveat, because a lot of coverage is running it the other way: several outlets frame this as a direct shot at a competitor's retention policy, but OpenAI's own page says nothing about competitors at all. Treat the rivalry angle as reported, not stated. What is solid is the direction. If you sell into health, legal or finance, “which model” is drifting from a benchmark question toward a procurement one.\x3C/p>\x3Cp>\x3Cstrong>A DeepSeek update, not a new story.\x3C/strong>\x3Cbr>We covered the V4-Pro pricing restructure on August 18, so this is one line rather than a section: since August 23, all day Saturday and Sunday count as off-peak. If you already moved batch work to dodge the peak windows, you can now park the heaviest weekly job on a weekend at half rate. That is the entire change.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#128581; Skip it\x3C/td>\x3C/tr>\x3C/table>\x3Cp>\x3Cstrong>Unitree's IPO, and the “the robots are here” week it produced.\x3C/strong> The stock debuted on Shanghai's STAR Market on August 18, popped 629% intraday and closed up 487%, landing near $53B. The underlying business earned roughly &yen;278M on &yen;1.7B of 2025 revenue and shipped about 5,500 humanoids. That is somewhere near 1,300 times earnings and 210 times sales. It was priced at about $9B on August 6, which means the entire re-rating happened in retail order books over twelve days, on an offer oversubscribed thousands of times over. Nothing about what the robots can do changed in that window. It is a genuinely interesting market story and there is nothing in it you can build on.\x3C/p>\x3Cp>The one thing worth doing this week takes about thirty seconds: search your codebase for \x3Cstrong>computer_20251124\x3C/strong>. That is the only item here whose request shape has already changed. Everything else is a header you can drop when convenient, or a version pin that will introduce itself later. Honest worst case if you ignore all of it - an agent that used to work starts throwing shape errors on a Tuesday, and you spend an hour finding out why.\x3C/p>\x3Cp>The 30-day challenge we opened on Friday is on Day 5, and it is built to be caught up on.\x3C/p>\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" align=\"center\" style=\"margin:30px auto 10px auto;\">\x3Ctr>\x3Ctd class=\"kit-cta\" align=\"center\" bgcolor=\"#0074CA\" style=\"border-radius:6px;\">\x3Ca href=\"https://www.skool.com/try-ai-academy/classroom/4bbc5d84?md=fd78ca15032d4d35bf36be42803c0705\" target=\"_blank\" rel=\"noopener noreferrer\">Start from Day 1 &rarr;\x3C/a>\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"text-align:center; margin:0 0 30px 0; font-size:13px; color:#5F5E5A;\">Free, no card. Day 1 takes 20 minutes.\x3C/p>\x3Chr>\x3Cp>\x3Cstrong>That's the Debrief.\x3C/strong>\x3C/p>\x3Cp>See you Friday for the Toolkit.\x3C/p>\x3Cp>From Drew and The AI Debrief team\x3C/p>\x3Chr>\x3Cp>\x3Cstrong>P.S.\x3C/strong> What's still pinned to a beta header inside something you own? Hit reply and tell us. We read every one.\x3C/p>\x3Cp>\x3Cstrong>P.P.S.\x3C/strong> New here, or skipped one? Every past edition is in the \x3Ca href=\"https://thedebrief.ai/archive\" target=\"_blank\" rel=\"noopener noreferrer\">archive\x3C/a>.\x3C/p>",edition:"debrief",issueNo:21},$R[75]={slug:"30-days-10-minutes-a-day-25496488",title:"30 days. 10 minutes a day.",subtitle:"This week's tool isn't an app. It's a plan &mdash; plus one exercise you can run today whether or not you take the rest of it.",date:"2026-08-21",thumbnail:"",blocks:$R[76]=[$R[77]={c:"The Toolkit &middot; No. 20 &middot; Fri Aug 21",t:"p"},$R[78]={c:"30 days. 10 minutes a day.",t:"p"},$R[79]={c:"This week's tool isn't an app. It's a plan &mdash; plus one exercise you can run today whether or not you take the rest of it.",t:"p"},$R[80]={c:"\x3Cstrong>Quick note before we start:\x3C/strong> you subscribed to The AI Debrief a while back, and then things went quiet on you. That's on me — we moved the whole newsletter to a new platform, and I didn't want to hit your inbox until it was worth it. This is the new era: the Debrief every Tuesday, the Toolkit every Friday. And if AI emails aren't your thing anymore, no hard feelings — the unsubscribe link at the bottom works on the first click.",t:"p"},$R[81]={c:"Welcome to the Toolkit, the Friday half of The AI Debrief. Tuesday tells you what happened; Friday hands you something to use. (If you missed it, \x3Ca href=\"https://thedebrief.ai/p/google-halved-the-price-until-january\" target=\"_blank\" rel=\"noopener noreferrer\">Tuesday's issue\x3C/a>
1 was about a model that's half price only until January.)",t:"p"},$R[82]={c:"Here's the problem this week's tool is built to solve, stated honestly. You read about something on Tuesday. You mean to try it. By Thursday it's buried, and the following week there's a new thing, and the cycle restarts. Six months of that and you know a hundred tool names while actually using about two. The gap isn't knowledge. It's reps &mdash; on a schedule, applied to your own work.",t:"p"},$R[83]={c:"In this email",t:"p"},$R[84]={c:"&raquo; The 30-day map &mdash; what each week actually covers",t:"p"},$R[85]={c:"&raquo; The Day 2 exercise, stripped down so you can run it in ten minutes today",t:"p"},$R[86]={c:"&raquo; Who should skip this, honestly",t:"p"},$R[87]={c:"The 30-Day AI Academy Challenge",t:"p"},$R[88]={c:"We opened it this week and it's free. Every day for thirty days you get three things: \x3Cstrong>one short lesson, one 10-minute worksheet, and one check-in.\x3C/strong> That's the entire commitment &mdash; no three-hour modules, no dashboard of unwatched videos.",t:"p"},$R[89]={c:"Runs for",t:"p"},$R[90]={c:"30 days",t:"p"},$R[91]={c:"Per day",t:"p"},$R[92]={c:"~10 min",t:"p"},$R[93]={c:"Cost",t:"p"},$R[94]={c:"$0",t:"p"},$R[95]={c:"The ten minutes is the feature, not a limitation. Anything that needs an hour gets postponed to a day you have an hour, and that day keeps not arriving. Ten minutes survives a bad week, which is the only thing that matters over a month.",t:"p"},$R[96]={c:"The worksheet is what makes it stick. A lesson you watch is entertainment. A lesson you apply to your own work the same day is a skill you still have in March.",t:"p"},$R[97]={c:"The month at a glance",t:"p"},$R[98]={c:"\x3Cstrong>Start Here (Days 1&ndash;2).\x3C/strong> A welcome and a tour, then your goal and your AI stack. Day 2 is the one that decides whether the rest lands &mdash; you pick what you're actually building toward, so every lesson after it has somewhere to go.",t:"p"},$R[99]={c:"\x3Cstrong>Week 1 &mdash; AI Foundations.\x3C/strong> How these models actually work and how to talk to them. Unglamorous, and the reason everything downstream runs on understanding instead of copy-paste.",t:"p"},$R[100]={c:"\x3Cstrong>Week 2 &mdash; Core Skills.\x3C/strong> The dense stretch: advanced prompting, custom instructions and memory, projects and knowledge bases, writing and content in your own voice, AI images and video, then data, docs and spreadsheets. It ends on \x3Cstrong>\x3Ca href=\"https://www.skool.com/try-ai-academy/classroom/4bbc5d84?md=573fd312a5f54b01bc5b046ecfd7e00b\" target=\"_blank\" rel=\"noopener noreferrer\">Day 16 with the capstone\x3C/a>\x3C/strong> &mdash; you assemble all of it into your own personal AI OS. Not a certificate. A working system, tuned to your job, that you keep.",t:"p"},$R[101]={c:"\x3Cstrong>Week 3 &mdash; Automations.\x3C/strong> The part almost nobody reaches on their own: getting the work to happen while you're not at the keyboard. This is where the ten minutes a day starts paying rent.",t:"p"},$R[102]={c:"\x3Cstrong>Week 4 &mdash; Build & Apply.\x3C/strong> The payoff stretch. You pick your path, build the AI system that actually fits it, and get into the ways people are turning these skills into income. Everything from the first three weeks points here &mdash; by the end you are not learning tools any more, you are running something.",t:"p"},$R[103]={c:"Here's the Day 2 exercise, stripped down. It works on paper, and it's the single highest-leverage thing in the first week.",t:"p"},$R[104]={c:"\x3Cstrong>1. List last week's five most repeated tasks.\x3C/strong> Not projects &mdash; tasks. \"Wrote the client update.\" \"Cleaned the export.\" \"Answered the same three questions.\" Be specific enough that someone else could do it from your description.",t:"p"},$R[105]={c:"\x3Cstrong>2. Put two numbers on each.\x3C/strong> Roughly how many hours it took, and how much you'd miss it on a scale of one to five. Most people find one task eating four-plus hours that they'd score a one.",t:"p"},$R[106]={c:"\x3Cstrong>3. That's your target.\x3C/strong> High hours, low attachment. Not the most interesting task &mdash; the most repeated one. Interesting work is where you want your hours going; repeated work is what you want to hand off first.",t:"p"},$R[107]={c:"\x3Cstrong>4. Name your stack, out loud.\x3C/strong> Which model you'll use, where the context lives, what \"done\" looks like. Most AI attempts fail here: people ask a chat window for something without having decided what a good answer looks like.",t:"p"},$R[108]={c:"Do those four steps and you'll get more out of the next month than most people get out of a course. The challenge just gives you twenty-nine more days of it, with the worksheet already written.",t:"p"},$R[109]={c:"Every 30-day anything has the same failure point, and it is not Day 1 or Day 30. It is Day 9 &mdash; far enough in that the novelty is gone, not far enough that you have seen a result yet. Three things get people through it.",t:"p"},$R[110]={c:"\x3Cstrong>Anchor it to something you already do.\x3C/strong> Not \"I will do it in the morning\" but \"I will do it while the coffee brews.\" A time anchored to an existing habit survives a busy week. A time anchored to intention does not.",t:"p"},$R[111]={c:"\x3Cstrong>Do the worksheet on real work.\x3C/strong> The temptation is to practise on a tidy made-up example. Don't. Use the actual half-finished thing sitting on your desk, however messy &mdash; messy is where the learning is, and it is the only version that transfers to Monday.",t:"p"},$R[112]={c:"\x3Cstrong>Miss one, never two.\x3C/strong> You will miss a day; everyone does. The rule that matters is that a missed day stays a missed day instead of becoming the end of the streak. Almost nobody quits because they skipped a Tuesday. They quit because skipping Tuesday made the whole thing feel already failed.",t:"p"},$R[113]={c:"If you want company for it, the challenge runs inside \x3Ca href=\"https://www.skool.com/try-ai-academy/about\" target=\"_blank\" rel=\"noopener noreferrer\">AI Academy\x3C/a>, where people post what they built each day. That daily check-in is the third piece, and it is the reason a group challenge finishes at a completely different rate than a course you bought alone.",t:"p"},$R[114]={c:"Why you care",t:"p"},$R[115]={c:"Ten minutes a day for a month beats a weekend binge you never revisit &mdash; consistency is the entire trick with this stuff. And you don't finish with a certificate, you finish with a system you still use. It costs nothing, Day 1 is a 20-minute welcome, and the worst case is you quit on Day 4 having already done the exercise above.",t:"p"},$R[116]={c:"\x3Cstrong>Who should skip it.\x3C/strong> If you're already running custom instructions, knowledge bases and automations daily, Weeks 1 and 2 will be review &mdash; start at Week 3 or skip this one entirely and I'll see you Tuesday. This is built for the much larger group who use AI most weeks, suspect they are using maybe a tenth of it, and have never had a reason to sit down and find out. Thirty days is long enough to prove that out and short enough that you will actually finish.",t:"p"},$R[117]={c:"Free, no card. Day 1 takes 20 minutes.",t:"p"},$R[118]={c:"\x3Cstrong>That's the Toolkit.\x3C/strong>",t:"p"},$R[119]={c:"See you Tuesday for the Debrief.",t:"p"},$R[120]={c:"From Drew and The AI Debrief team",t:"p"},$R[121]={c:"\x3Cstrong>P.S.\x3C/strong> If you start the challenge, hit reply and tell me the goal you pick on Day 2. I read every one, and I'll be checking in on the challengers all month.",t:"p"},$R[122]={c:"\x3Cstrong>P.P.S.\x3C/strong> New here, or skipped one? Every past edition is in the \x3Ca href=\"https://thedebrief.ai/archive\" target=\"_blank\" rel=\"noopener noreferrer\">archive\x3C/a>.",t:"p"}],edition:"toolkit",issueNo:null},$R[123]={slug:"30-days-10-minutes-a-day",title:"30 days. 10 minutes a day.",subtitle:"This week's tool isn't an app. It's a plan &mdash; plus one exercise you can run today whether or not you take the rest of it.",date:"2026-08-21",thumbnail:"https://embed.filekitcdn.com/e/jeCKAoLJGmzsPEwMDzFkXA/2JnnkyFzrp9CnCrRRPMo11",blocks:$R[124]=[],html:"\x3Ctable cellPadding=\"0\" cellSpacing=\"0\" style=\"width:100%;margin:0 auto\">\x3Ctbody>\x3Ctr>\x3Ctd>\x3Cdiv>\x3Ctable cellPadding=\"0\" cellSpacing=\"0\" style=\"width:100%;margin:0 auto\">\x3Ctbody>\x3Ctr>\x3Ctd>\x3Cdiv>\x3Cp>Welcome to the Toolkit, the Friday half of The AI Debrief. Tuesday tells you what happened; Friday hands you something to use. (If you missed it, \x3Ca href=\"https://thedebrief.ai/p/google-halved-the-price-until-january\" target=\"_blank\" rel=\"noopener noreferrer\">Tuesday's issue\x3C/a>
1 was about a model that's half price only until January.)\x3C/p>\x3Cp>Here's the problem this week's tool is built to solve, stated honestly. You read about something on Tuesday. You mean to try it. By Thursday it's buried, and the following week there's a new thing, and the cycle restarts. Six months of that and you know a hundred tool names while actually using about two. The gap isn't knowledge. It's reps &mdash; on a schedule, applied to your own work.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 8px 0;\">\x3Ctr>\x3Ctd style=\"border-top:1px solid #E5E5E5; border-bottom:1px solid #E5E5E5; padding:14px 2px;\">\x3Cp style=\"margin:0 0 8px 0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#5F5E5A;\">In this email\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; The 30-day map &mdash; what each week actually covers\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; The Day 2 exercise, stripped down so you can run it in ten minutes today\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.5; color:#0A0A0A;\">&raquo; Who should skip this, honestly\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#129520; The tool: a month with a map\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"margin:16px 0 8px 0; font-size:19px; font-weight:700; line-height:1.3; color:#0A0A0A;\">The 30-Day AI Academy Challenge\x3C/p>\x3Cp>We opened it this week and it's free. Every day for thirty days you get three things: \x3Cstrong>one short lesson, one 10-minute worksheet, and one check-in.\x3C/strong> That's the entire commitment &mdash; no three-hour modules, no dashboard of unwatched videos.\x3C/p>\x3Ctable class=\"kit-stack\" role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:22px 0 22px 0;\">\x3Ctr>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#F6F6F4; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#5F5E5A;\">Runs for\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0A0A0A;\">30 days\x3C/p>\x3C/td>\x3Ctd width=\"2%\">&nbsp;\x3C/td>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#E6F1FB; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#0074CA;\">Per day\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0074CA;\">~10 min\x3C/p>\x3C/td>\x3Ctd width=\"2%\">&nbsp;\x3C/td>\x3Ctd width=\"32%\" valign=\"top\" style=\"background-color:#F6F6F4; padding:14px 12px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#5F5E5A;\">Cost\x3C/p>\x3Cp style=\"margin:4px 0 0 0; font-size:24px; font-weight:700; color:#0A0A0A;\">$0\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Cp>The ten minutes is the feature, not a limitation. Anything that needs an hour gets postponed to a day you have an hour, and that day keeps not arriving. Ten minutes survives a bad week, which is the only thing that matters over a month.\x3C/p>\x3Cp>The worksheet is what makes it stick. A lesson you watch is entertainment. A lesson you apply to your own work the same day is a skill you still have in March.\x3C/p>\x3Cp style=\"margin:26px 0 10px 0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#5F5E5A;\">The month at a glance\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 12px 0; border-collapse:separate; border-spacing:3px;\">\x3Ctr>\x3Ctd class=\"kit-code\" width=\"16.6%\" align=\"center\" style=\"background-color:#0A0A0A; padding:11px 0; font-size:13px; font-weight:700; color:#FFFFFF;\">1\x3C/td>\x3Ctd class=\"kit-code\" width=\"16.6%\" align=\"center\" style=\"background-color:#0A0A0A; padding:11px 0; font-size:13px; font-weight:700; color:#FFFFFF;\">2\x3C/td>\x3Ctd width=\"16.6%\" align=\"center\" style=\"background-color:#E6F1FB; padding:11px 0; font-size:13px; font-weight:700; color:#0074CA;\">3\x3C/td>\x3Ctd width=\"16.6%\" align=\"center\" style=\"background-color:#E6F1FB; padding:11px 0; font-size:13px; font-weight:700; color:#0074CA;\">4\x3C/td>\x3Ctd width=\"16.6%\" align=\"center\" style=\"background-color:#E6F1FB; padding:11px 0; font-size:13px; font-weight:700; color:#0074CA;\">5\x3C/td>\x3Ctd width=\"16.6%\" align=\"center\" style=\"background-color:#E6F1FB; padding:11px 0; font-size:13px; font-weight:700; color:#0074CA;\">6\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd align=\"center\" style=\"background-color:#E6F1FB; padding:11px 0; font-size:13px; font-weight:700; color:#0074CA;\">7\x3C/td>\x3Ctd align=\"center\" style=\"background-color:#E6F1FB; padding:11px 0; font-size:13px; font-weight:700; color:#0074CA;\">8\x3C/td>\x3Ctd align=\"center\" style=\"background-color:#E6F1FB; padding:11px 0; font-size:13px; font-weight:700; color:#0074CA;\">9\x3C/td>\x3Ctd align=\"center\" style=\"background-color:#CFE6FA; padding:11px 0; font-size:13px; font-weight:700; color:#0A4A7A;\">10\x3C/td>\x3Ctd align=\"center\" style=\"background-color:#CFE6FA; padding:11px 0; font-size:13px; font-weight:700; color:#0A4A7A;\">11\x3C/td>\x3Ctd align=\"center\" style=\"background-color:#CFE6FA; padding:11px 0; font-size:13px; font-weight:700; color:#0A4A7A;\">12\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd align=\"center\" style=\"background-color:#CFE6FA; padding:11px 0; font-size:13px; font-weight:700; color:#0A4A7A;\">13\x3C/td>\x3Ctd align=\"center\" style=\"background-color:#CFE6FA; padding:11px 0; font-size:13px; font-weight:700; color:#0A4A7A;\">14\x3C/td>\x3Ctd align=\"center\" style=\"background-color:#CFE6FA; padding:11px 0; font-size:13px; font-weight:700; color:#0A4A7A;\">15\x3C/td>\x3Ctd align=\"center\" style=\"background-color:#00BC7D; padding:11px 0; font-size:13px; font-weight:700; color:#FFFFFF;\">16\x3C/td>\x3Ctd align=\"center\" style=\"background-color:#0074CA; padding:11px 0; font-size:13px; font-weight:700; color:#FFFFFF;\">17\x3C/td>\x3Ctd align=\"center\" style=\"background-color:#0074CA; padding:11px 0; font-size:13px; font-weight:700; color:#FFFFFF;\">18\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd align=\"center\" style=\"background-color:#0074CA; padding:11px 0; font-size:13px; font-weight:700; color:#FFFFFF;\">19\x3C/td>\x3Ctd align=\"center\" style=\"background-color:#0074CA; padding:11px 0; font-size:13px; font-weight:700; color:#FFFFFF;\">20\x3C/td>\x3Ctd align=\"center\" style=\"background-color:#0074CA; padding:11px 0; font-size:13px; font-weight:700; color:#FFFFFF;\">21\x3C/td>\x3Ctd align=\"center\" style=\"background-color:#0074CA; padding:11px 0; font-size:13px; font-weight:700; color:#FFFFFF;\">22\x3C/td>\x3Ctd align=\"center\" style=\"background-color:#0074CA; padding:11px 0; font-size:13px; font-weight:700; color:#FFFFFF;\">23\x3C/td>\x3Ctd class=\"kit-code\" align=\"center\" style=\"background-color:#0A4A7A; padding:11px 0; font-size:13px; font-weight:700; color:#FFFFFF;\">24\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd class=\"kit-code\" align=\"center\" style=\"background-color:#0A4A7A; padding:11px 0; font-size:13px; font-weight:700; color:#FFFFFF;\">25\x3C/td>\x3Ctd class=\"kit-code\" align=\"center\" style=\"background-color:#0A4A7A; padding:11px 0; font-size:13px; font-weight:700; color:#FFFFFF;\">26\x3C/td>\x3Ctd class=\"kit-code\" align=\"center\" style=\"background-color:#0A4A7A; padding:11px 0; font-size:13px; font-weight:700; color:#FFFFFF;\">27\x3C/td>\x3Ctd class=\"kit-code\" align=\"center\" style=\"background-color:#0A4A7A; padding:11px 0; font-size:13px; font-weight:700; color:#FFFFFF;\">28\x3C/td>\x3Ctd class=\"kit-code\" align=\"center\" style=\"background-color:#0A4A7A; padding:11px 0; font-size:13px; font-weight:700; color:#FFFFFF;\">29\x3C/td>\x3Ctd class=\"kit-code\" align=\"center\" style=\"background-color:#0A4A7A; padding:11px 0; font-size:13px; font-weight:700; color:#FFFFFF;\">30\x3C/td>\x3C/tr>\x3C/table>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 24px 0;\">\x3Ctr>\x3Ctd style=\"padding:2px 0; font-size:12px; line-height:1.9; color:#5F5E5A;\">\x3Cspan style=\"background-color:#0A0A0A; color:#FFFFFF; padding:2px 7px; font-weight:700;\">1&ndash;2\x3C/span> Start Here &nbsp;&middot;&nbsp; \x3Cspan style=\"background-color:#E6F1FB; color:#0074CA; padding:2px 7px; font-weight:700;\">3&ndash;
19\x3C/span> Foundations &nbsp;&middot;&nbsp; \x3Cspan style=\"background-color:#CFE6FA; color:#0A4A7A; padding:2px 7px; font-weight:700;\">10&ndash;16\x3C/span> Core Skills &nbsp;&middot;&nbsp; \x3Cspan style=\"background-color:#00BC7D; color:#FFFFFF; padding:2px 7px; font-weight:700;\">16\x3C/span> Capstone &nbsp;&middot;&nbsp; \x3Cspan style=\"background-color:#0074CA; color:#FFFFFF; padding:2px 7px; font-weight:700;\">17&ndash;23\x3C/span> Automations &nbsp;&middot;&nbsp; \x3Cspan style=\"background-color:#0A4A7A; color:#FFFFFF; padding:2px 7px; font-weight:700;\">24&ndash;30\x3C/span> Build &amp; Apply\x3C/td>\x3C/tr>\x3C/table>\x3Cp>\x3Cstrong>Start Here (Days 1&ndash;2).\x3C/strong> A welcome and a tour, then your goal and your AI stack. Day 2 is the one that decides whether the rest lands &mdash; you pick what you're actually building toward, so every lesson after it has somewhere to go.\x3C/p>\x3Cp>\x3Cstrong>Week 1 &mdash; AI Foundations.\x3C/strong> How these models actually work and how to talk to them. Unglamorous, and the reason everything downstream runs on understanding instead of copy-paste.\x3C/p>\x3Cp>\x3Cstrong>Week 2 &mdash; Core Skills.\x3C/strong> The dense stretch: advanced prompting, custom instructions and memory, projects and knowledge bases, writing and content in your own voice, AI images and video, then data, docs and spreadsheets. It ends on \x3Cstrong>\x3Ca href=\"https://www.skool.com/try-ai-academy/classroom/4bbc5d84?md=573fd312a5f54b01bc5b046ecfd7e00b\" target=\"_blank\" rel=\"noopener noreferrer\">Day 16 with the capstone\x3C/a>\x3C/strong> &mdash; you assemble all of it into your own personal AI OS. Not a certificate. A working system, tuned to your job, that you keep.\x3C/p>\x3Cp>\x3Cstrong>Week 3 &mdash; Automations.\x3C/strong> The part almost nobody reaches on their own: getting the work to happen while you're not at the keyboard. This is where the ten minutes a day starts paying rent.\x3C/p>\x3Cp>\x3Cstrong>Week 4 &mdash; Build &amp; Apply.\x3C/strong> The payoff stretch. You pick your path, build the AI system that actually fits it, and get into the ways people are turning these skills into income. Everything from the first three weeks points here &mdash; by the end you are not learning tools any more, you are running something.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#9889; Try this today &mdash; ten minutes, no signup\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"margin-top:16px;\">Here's the Day 2 exercise, stripped down. It works on paper, and it's the single highest-leverage thing in the first week.\x3C/p>\x3Cp>\x3Cstrong>1. List last week's five most repeated tasks.\x3C/strong> Not projects &mdash; tasks. \"Wrote the client update.\" \"Cleaned the export.\" \"Answered the same three questions.\" Be specific enough that someone else could do it from your description.\x3C/p>\x3Cp>\x3Cstrong>2. Put two numbers on each.\x3C/strong> Roughly how many hours it took, and how much you'd miss it on a scale of one to five. Most people find one task eating four-plus hours that they'd score a one.\x3C/p>\x3Cp>\x3Cstrong>3. That's your target.\x3C/strong> High hours, low attachment. Not the most interesting task &mdash; the most repeated one. Interesting work is where you want your hours going; repeated work is what you want to hand off first.\x3C/p>\x3Cp>\x3Cstrong>4. Name your stack, out loud.\x3C/strong> Which model you'll use, where the context lives, what \"done\" looks like. Most AI attempts fail here: people ask a chat window for something without having decided what a good answer looks like.\x3C/p>\x3Cp>Do those four steps and you'll get more out of the next month than most people get out of a course. The challenge just gives you twenty-nine more days of it, with the worksheet already written.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">&#128681; How not to quit on Day 9\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"margin-top:16px;\">Every 30-day anything has the same failure point, and it is not Day 1 or Day 30. It is Day 9 &mdash; far enough in that the novelty is gone, not far enough that you have seen a result yet. Three things get people through it.\x3C/p>\x3Cp>\x3Cstrong>Anchor it to something you already do.\x3C/strong> Not \"I will do it in the morning\" but \"I will do it while the coffee brews.\" A time anchored to an existing habit survives a busy week. A time anchored to intention does not.\x3C/p>\x3Cp>\x3Cstrong>Do the worksheet on real work.\x3C/strong> The temptation is to practise on a tidy made-up example. Don't. Use the actual half-finished thing sitting on your desk, however messy &mdash; messy is where the learning is, and it is the only version that transfers to Monday.\x3C/p>\x3Cp>\x3Cstrong>Miss one, never two.\x3C/strong> You will miss a day; everyone does. The rule that matters is that a missed day stays a missed day instead of becoming the end of the streak. Almost nobody quits because they skipped a Tuesday. They quit because skipping Tuesday made the whole thing feel already failed.\x3C/p>\x3Cp>If you want company for it, the challenge runs inside \x3Ca href=\"https://www.skool.com/try-ai-academy/about\" target=\"_blank\" rel=\"noopener noreferrer\">AI Academy\x3C/a>, where people post what they built each day. That daily check-in is the third piece, and it is the reason a group challenge finishes at a completely different rate than a course you bought alone.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 18px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #0A0A0A; padding:14px 16px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0A0A0A;\">Why you care\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:15px; line-height:1.6; color:#0A0A0A;\">Ten minutes a day for a month beats a weekend binge you never revisit &mdash; consistency is the entire trick with this stuff. And you don't finish with a certificate, you finish with a system you still use. It costs nothing, Day 1 is a 20-minute welcome, and the worst case is you quit on Day 4 having already done the exercise above.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Cp>\x3Cstrong>Who should skip it.\x3C/strong> If you're already running custom instructions, knowledge bases and automations daily, Weeks 1 and 2 will be review &mdash; start at Week 3 or skip this one entirely and I'll see you Tuesday. This is built for the much larger group who use AI most weeks, suspect they are using maybe a tenth of it, and have never had a reason to sit down and find out. Thirty days is long enough to prove that out and short enough that you will actually finish.\x3C/p>\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" align=\"center\" style=\"margin:30px auto 10px auto;\">\x3Ctr>\x3Ctd class=\"kit-cta\" align=\"center\" bgcolor=\"#0074CA\" style=\"border-radius:6px;\">\x3Ca href=\"https://www.skool.com/try-ai-academy/classroom/4bbc5d84?md=fd78ca15032d4d35bf36be42803c0705\" target=\"_blank\" rel=\"noopener noreferrer\">Start Day 1 &mdash; free &rarr;\x3C/a>\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"text-align:center; margin:0 0 30px 0; font-size:13px; color:#5F5E5A;\">Free, no card. Day 1 takes 20 minutes.\x3C/p>\x3Chr>\x3Cp>\x3Cstrong>That's the Toolkit.\x3C/strong>\x3C/p>\x3Cp>See you Tuesday for the Debrief.\x3C/p>\x3Cp>From Drew and The AI Debrief team\x3C/p>\x3Chr>\x3Cp>\x3Cstrong>P.S.\x3C/strong> If you start the challenge, hit reply and tell me the goal you pick on Day 2. I read every one, and I'll be checking in on the challengers all month.\x3C/p>\x3Cp>\x3Cstrong>P.P.S.\x3C/strong> New here, or skipped one? Every past edition is in the \x3Ca href=\"https://thedebrief.ai/archive\" target=\"_blank\" rel=\"noopener noreferrer\">archive\x3C/a>.\x3C/p>\x3C/div>\x3C/td>\x3C/tr>\x3C/tbody>\x3C/table>\x3C/div>\x3C/td>\x3C/tr>\x3C/tbody>\x3C/table>",edition:"toolkit",issueNo:20},$R[125]={slug:"google-halved-the-price-until-january-25495572",title:"Google halved the price. Until January.",subt
1itle:"The intro rate on Gemini 3.7 Flash doubles on January 1, and almost nobody covering it said so.",date:"2026-08-18",thumbnail:"",blocks:$R[126]=[$R[127]={c:"The Debrief &middot; No. 19 &middot; Tue Aug 18",t:"p"},$R[128]={c:"Google halved the price. Until January.",t:"p"},$R[129]={c:"The intro rate on Gemini 3.7 Flash doubles on January 1, and almost nobody covering it said so.",t:"p"},$R[130]={c:"\x3Cstrong>Quick note before we start:\x3C/strong> you subscribed to The AI Debrief a while back, and then things went quiet on you. That's on me — we moved the whole newsletter to a new platform, and I didn't want to hit your inbox until it was worth it. This is the new era: the Debrief every Tuesday, the Toolkit every Friday. And if AI emails aren't your thing anymore, no hard feelings — the unsubscribe link at the bottom works on the first click.",t:"p"},$R[131]={c:"Welcome to the Debrief, the Tuesday half of The AI Debrief. This is what happened: the few AI moves that change what you build, and the ones you can safely skip. Friday is the Toolkit, where we hand you something to use.",t:"p"},$R[132]={c:"Here's what we're getting into this week. Two price changes, one of which already hit your bill on Saturday, and a set of endpoints that stopped working this morning.",t:"p"},$R[133]={c:"In this email",t:"p"},$R[134]={c:"» The half-price model that isn't, and what doubles on January 1",t:"p"},$R[135]={c:"» The image endpoints that died today, and the one-line fix",t:"p"},$R[136]={c:"» Why your DeepSeek bill went up on Saturday without an email",t:"p"},$R[137]={c:"Gemini 3.7 Flash is cheap on a timer",t:"p"},$R[138]={c:"Google shipped Gemini 3.7 Flash on August 13, three weeks after 3.6 Flash and, oddly, before Gemini 3.5 Pro. Input runs $0.75 per million tokens, output $3.75. Most of the coverage called it a price cut and moved on.",t:"p"},$R[139]={c:"Today",t:"p"},$R[140]={c:"$0.75",t:"p"},$R[141]={c:"January 1",t:"p"},$R[142]={c:"$1.50",t:"p"},$R[143]={c:"Change",t:"p"},$R[144]={c:"+100%",t:"p"},$R[145]={c:"Read the pricing page and the word \"introductory\" is doing a lot of work. Every number on it is dated through December 31, 2026. On January 1 input goes to $1.50 and output to $7.50. Batch pricing doubles too, from $0.375 and $1.875 to $0.75 and $3.75. Context caching doubles: reads from $0.075 to $0.15, and storage from $0.50 to $1.00 per million tokens per hour. Nothing is grandfathered, and the increase is automatic.",t:"p"},$R[146]={c:"The second thing worth noticing is the order Google shipped in. A Flash model before its Pro model, with the launch pointed squarely at coding and agents. On Google's own AutomationBench, which measures enterprise workflow automation, 3.7 Flash scores 30.4% against 17.0% for its predecessor. That is close to a doubling. It is also a vendor-reported number, so treat it as a claim rather than a fact until somebody independent replicates it.",t:"p"},$R[147]={c:"I keep coming back to the calendar rather than the benchmark. Four and a half months at half price is real money if you are running multi-step agents, and it is also exactly long enough to build a cost model, forget how you built it, and get surprised in January. The pattern underneath is the one to actually watch: Flash-tier models absorbing work that used to require a frontier tier. If that AutomationBench jump survives independent testing, it reprices a whole category of automation, and the interesting question stops being which model is smartest.",t:"p"},$R[148]={c:"Why you care",t:"p"},$R[149]={c:"If you are running agent or coding workloads on a frontier model, there is a real arbitrage here until December 31, and you should take it. Then put a reminder on your calendar for December 15 to re-run your numbers at $1.50 and $7.50, because that is the price you will actually be paying. A cost model with an undated assumption in it is not a cost model.",t:"p"},$R[150]={c:"\x3Ca href=\"https://ai.google.dev/gemini-api/docs/pricing\" target=\"_blank\" rel=\"noopener noreferrer\">See the pricing page →\x3C/a>",t:"p"},$R[151]={c:"→ \x3Cstrong>Google shut off the Imagen 4 endpoints today.\x3C/strong> All three variants - generate, ultra and fast - hit end of life on August 17. The replacement is gemini-3.1-flash-image. If you have an image pipeline calling those endpoints, it is not breaking soon. It broke this morning. One string swap. The robotics preview endpoint goes on August 31, so check that too while you are in there.",t:"p"},$R[152]={c:"→ \x3Cstrong>Your DeepSeek bill went up on Saturday.\x3C/strong> V4-Pro launched August 13 and the restructured API pricing took effect August 16 at 16:00 UTC. Outlets reported that date at least three different ways, so if you read somewhere that prices \"change this week,\" they already changed. The detail nobody led with is the useful one: off-peak is exactly half of peak, and peak runs 01:00 to 04:00 and 06:00 to 10:00 UTC. Batch work is now schedulable for a straight 50% discount. Caixin reports some rates rising as much as 1,100%; I could not reproduce that from the published rate card, so treat it as reported rather than verified.",t:"p"},$R[153]={c:"→ \x3Cstrong>Six companies that agree on nothing agreed on a plugin format.\x3C/strong> Agent Plugins 1.0 shipped with AWS, Microsoft, OpenAI, Google, Vercel and Anysphere as co-maintainers. It bundles agent skills and MCP servers into a single installable unit, with the vendor-specific parts namespaced so a plugin stays portable across tools. If you build internal agent tooling, this is the first credible sign you get to build it once instead of once per platform.",t:"p"},$R[154]={c:"Nvidia backstopping $105 billion for OpenAI's 10-gigawatt data center in Ohio. Biggest number of the week and the least relevant. You cannot buy it, use it, or build on it, and the capacity lands years out. The money is real; the relevance is zero.",t:"p"},$R[155]={c:"One in the same bucket that deserves a caveat rather than a shrug: the Stripe and OpenRouter $7 billion acquisition headline. Bloomberg reported it. Neither Stripe's newsroom nor OpenRouter's own blog has said a word, and OpenRouter published twice yesterday about something else entirely. Several outlets have already quietly upgraded \"reportedly\" to \"closes.\" If OpenRouter is your routing layer, change nothing yet.",t:"p"},$R[156]={c:"The one thing actually worth doing this week: check whether anything you own calls an Imagen 4 endpoint. That is a live break, not a warning. Everything else here is a calendar reminder. Honest worst case if you ignore all of it - you overpay a bit in January, and you find out about the broken image job when a customer tells you.",t:"p"},$R[157]={c:"One more thing before you go: we just opened the \x3Cstrong>free 30-Day AI Academy Challenge\x3C/strong> — one short lesson, one 10-minute worksheet, and one check-in a day, for a month. By Day 16 you've built your own personal AI OS. Day 1 is live now, and it costs nothing.",t:"p"},$R[158]={c:"\x3Ca href=\"https://www.skool.com/try-ai-academy/classroom/4bbc5d84?md=fd78ca15032d4d35bf36be42803c0705\" target=\"_blank\" rel=\"noopener noreferrer\">Start Day 1 →\x3C/a>",t:"p"},$R[159]={c:"\x3Cstrong>That's the Debrief.\x3C/strong>",t:"p"},$R[160]={c:"See you Friday for the Toolkit.",t:"p"},$R[161]={c:"From Drew and The AI Debrief team",t:"p"},$R[162]={c:"\x3Cstrong>P.S.\x3C/strong> What's the oldest API endpoint still running inside something you own? Hit reply and tell us. We read every one.",t:"p"},$R[163]={c:"\x3Cstrong>P.P.S.\x3C/strong> New here, or skipped one? Every past edition is in the \x3Ca href=\"https://thedebrief.ai/archive\" target=\"_blank\" rel=\"noopener noreferrer\">archive\x3C/a>.",t:"p"}],edition:"debrief",issueNo:null},$R[164]={slug:"google-halved-the-price-until-january",title:"Google halved the price. Until January.",subt
1itle:"The intro rate on Gemini 3.7 Flash doubles on January 1, and almost nobody covering it said so.",date:"2026-08-18",thumbnail:"https://embed.filekitcdn.com/e/jeCKAoLJGmzsPEwMDzFkXA/a82NHHfS3yD23WUKdWBva1",blocks:$R[165]=[],html:"\x3Ctable cellPadding=\"0\" cellSpacing=\"0\" style=\"width:100%;margin:0 auto\">\x3Ctbody>\x3Ctr>\x3Ctd>\x3Cdiv>\x3Ctable cellPadding=\"0\" cellSpacing=\"0\" style=\"width:100%;margin:0 auto\">\x3Ctbody>\x3Ctr>\x3Ctd>\x3Cdiv>\x3Cp>Welcome to the Debrief, the Tuesday half of The AI Debrief. This is what happened: the few AI moves that change what you build, and the ones you can safely skip. Friday is the Toolkit, where we hand you something to use.\x3C/p>\x3Cp>Here's what we're getting into this week. Two price changes, one of which already hit your bill on Saturday, and a set of endpoints that stopped working this morning.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 8px 0;\">\x3Ctr>\x3Ctd style=\"border-top:1px solid #E5E5E5; border-bottom:1px solid #E5E5E5; padding:14px 2px;\">\x3Cp style=\"margin:0 0 8px 0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#5F5E5A;\">In this email\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">» The half-price model that isn't, and what doubles on January 1\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">» The image endpoints that died today, and the one-line fix\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.5; color:#0A0A0A;\">» Why your DeepSeek bill went up on Saturday without an email\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">💸 The discount has an expiration date\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"margin:16px 0 8px 0; font-size:19px; font-weight:700; line-height:1.3; color:#0A0A0A;\">Gemini 3.7 Flash is cheap on a timer\x3C/p>\x3Cp>Google shipped Gemini 3.7 Flash on August 13, three weeks after 3.6 Flash and, oddly, before Gemini 3.5 Pro. Input runs $0.75 per million tokens, output $3.75. Most of the coverage called it a price cut and moved on.\x3C/p>\x3Ctable class=\"kit-stack\" role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 18px 0;\">\x3Ctr>\x3Ctd width=\"33%\" valign=\"top\" style=\"background-color:#F6F6F4; padding:12px 14px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#5F5E5A;\">Today\x3C/p>\x3Cp style=\"margin:2px 0 0 0; font-size:20px; font-weight:700; color:#5F5E5A;\">$0.75\x3C/p>\x3C/td>\x3Ctd width=\"8\">&nbsp;\x3C/td>\x3Ctd width=\"33%\" valign=\"top\" style=\"background-color:#E6F1FB; padding:12px 14px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#0074CA;\">January 1\x3C/p>\x3Cp style=\"margin:2px 0 0 0; font-size:20px; font-weight:700; color:#0074CA;\">$1.50\x3C/p>\x3C/td>\x3Ctd width=\"8\">&nbsp;\x3C/td>\x3Ctd width=\"33%\" valign=\"top\" style=\"background-color:#F6F6F4; padding:12px 14px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#5F5E5A;\">Change\x3C/p>\x3Cp style=\"margin:2px 0 0 0; font-size:20px; font-weight:700; color:#0A0A0A;\">+100%\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Cp>Read the pricing page and the word \"introductory\" is doing a lot of work. Every number on it is dated through December 31, 2026. On January 1 input goes to $1.50 and output to $7.50. Batch pricing doubles too, from $0.375 and $1.875 to $0.75 and $3.75. Context caching doubles: reads from $0.075 to $0.15, and storage from $0.50 to $1.00 per million tokens per hour. Nothing is grandfathered, and the increase is automatic.\x3C/p>\x3Cp>The second thing worth noticing is the order Google shipped in. A Flash model before its Pro model, with the launch pointed squarely at coding and agents. On Google's own AutomationBench, which measures enterprise workflow automation, 3.7 Flash scores 30.4% against 17.0% for its predecessor. That is close to a doubling. It is also a vendor-reported number, so treat it as a claim rather than a fact until somebody independent replicates it.\x3C/p>\x3Cp>I keep coming back to the calendar rather than the benchmark. Four and a half months at half price is real money if you are running multi-step agents, and it is also exactly long enough to build a cost model, forget how you built it, and get surprised in January. The pattern underneath is the one to actually watch: Flash-tier models absorbing work that used to require a frontier tier. If that AutomationBench jump survives independent testing, it reprices a whole category of automation, and the interesting question stops being which model is smartest.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 18px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #0A0A0A; padding:14px 16px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0A0A0A;\">Why you care\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:15px; line-height:1.6; color:#0A0A0A;\">If you are running agent or coding workloads on a frontier model, there is a real arbitrage here until December 31, and you should take it. Then put a reminder on your calendar for December 15 to re-run your numbers at $1.50 and $7.50, because that is the price you will actually be paying. A cost model with an undated assumption in it is not a cost model.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Cp>\x3Ca href=\"https://ai.google.dev/gemini-api/docs/pricing\" target=\"_blank\" rel=\"noopener noreferrer\">See the pricing page →\x3C/a>\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">
1⚡ Worth your time\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"margin-top:16px;\">→ \x3Cstrong>Google shut off the Imagen 4 endpoints today.\x3C/strong> All three variants - generate, ultra and fast - hit end of life on August 17. The replacement is gemini-3.1-flash-image. If you have an image pipeline calling those endpoints, it is not breaking soon. It broke this morning. One string swap. The robotics preview endpoint goes on August 31, so check that too while you are in there.\x3C/p>\x3Cp>→ \x3Cstrong>Your DeepSeek bill went up on Saturday.\x3C/strong> V4-Pro launched August 13 and the restructured API pricing took effect August 16 at 16:00 UTC. Outlets reported that date at least three different ways, so if you read somewhere that prices \"change this week,\" they already changed. The detail nobody led with is the useful one: off-peak is exactly half of peak, and peak runs 01:00 to 04:00 and 06:00 to 10:00 UTC. Batch work is now schedulable for a straight 50% discount. Caixin reports some rates rising as much as 1,100%; I could not reproduce that from the published rate card, so treat it as reported rather than verified.\x3C/p>\x3Cp>→ \x3Cstrong>Six companies that agree on nothing agreed on a plugin format.\x3C/strong> Agent Plugins 1.0 shipped with AWS, Microsoft, OpenAI, Google, Vercel and Anysphere as co-maintainers. It bundles agent skills and MCP servers into a single installable unit, with the vendor-specific parts namespaced so a plugin stays portable across tools. If you build internal agent tooling, this is the first credible sign you get to build it once instead of once per platform.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">🙅 Skip it\x3C/td>\x3C/tr>\x3C/table>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:16px 0 18px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #00BC7D; padding:16px 18px;\">\x3Cp style=\"margin:0 0 10px 0; font-size:15px; line-height:1.6; color:#0A0A0A;\">Nvidia backstopping $105 billion for OpenAI's 10-gigawatt data center in Ohio. Biggest number of the week and the least relevant. You cannot buy it, use it, or build on it, and the capacity lands years out. The money is real; the relevance is zero.\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.6; color:#0A0A0A;\">One in the same bucket that deserves a caveat rather than a shrug: the Stripe and OpenRouter $7 billion acquisition headline. Bloomberg reported it. Neither Stripe's newsroom nor OpenRouter's own blog has said a word, and OpenRouter published twice yesterday about something else entirely. Several outlets have already quietly upgraded \"reportedly\" to \"closes.\" If OpenRouter is your routing layer, change nothing yet.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Chr>\x3Cp>The one thing actually worth doing this week: check whether anything you own calls an Imagen 4 endpoint. That is a live break, not a warning. Everything else here is a calendar reminder. Honest worst case if you ignore all of it - you overpay a bit in January, and you find out about the broken image job when a customer tells you.\x3C/p>\x3Cp>One more thing before you go: we just opened the \x3Cstrong>free 30-Day AI Academy Challenge\x3C/strong> — one short lesson, one 10-minute worksheet, and one check-in a day, for a month. By Day 16 you've built your own personal AI OS. Day 1 is live now, and it costs nothing.\x3C/p>\x3Cp>\x3Ca href=\"https://www.skool.com/try-ai-academy/classroom/4bbc5d84?md=fd78ca15032d4d35bf36be42803c0705\" target=\"_blank\" rel=\"noopener noreferrer\">Start Day 1 →\x3C/a>\x3C/p>\x3Cp>\x3Cstrong>That's the Debrief.\x3C/strong>\x3C/p>\x3Cp>See you Friday for the Toolkit.\x3C/p>\x3Cp>From Drew and The AI Debrief team\x3C/p>\x3Chr>\x3Cp>\x3Cstrong>P.S.\x3C/strong> What's the oldest API endpoint still running inside something you own? Hit reply and tell us. We read every one.\x3C/p>\x3Cp>\x3Cstrong>P.P.S.\x3C/strong> New here, or skipped one? Every past edition is in the \x3Ca href=\"https://thedebrief.ai/archive\" target=\"_blank\" rel=\"noopener noreferrer\">archive\x3C/a>.\x3C/p>\x3C/div>\x3C/td>\x3C/tr>\x3C/tbody>\x3C/table>\x3C/div>\x3C/td>\x3C/tr>\x3C/tbody>\x3C/table>",edition:"debrief",issueNo:19},$R[166]={slug:"the-100-agent-is-now-20",title:"The $100 agent is now $20",subtitle:"The price barrier fell. The permissions barrier didn't move an inch.",date:"2026-08-14",thumbnail:"https://embed.filekitcdn.com/e/jeCKAoLJGmzsPEwMDzFkXA/tntaQfaw9557rewKNQRtBF",blocks:$R[167]=[],html:"\x3Ctable cellPadding=\"0\" cellSpacing=\"0\" style=\"width:100%;margin:0 auto\">\x3Ctbody>\x3Ctr>\x3Ctd>\x3Cdiv>\x3Cp>Welcome to the Toolkit, the Friday half of The AI Debrief. Tuesday is what happened. Friday is what to do about it: one tool worth your time, one play you can run today, and nothing you have to buy.\x3C/p>\x3Cp>Here's what we're getting into this week. The agent that runs your inbox while you sleep just dropped to $20 a month, the cat
1ch is who it's allowed to work for, and the fix for both is a page you write in twenty minutes.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 8px 0;\">\x3Ctr>\x3Ctd style=\"border-top:1px solid #E5E5E5; border-bottom:1px solid #E5E5E5; padding:14px 2px;\">\x3Cp style=\"margin:0 0 8px 0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#5F5E5A;\">In this email\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">» The always-on agent is $20 now, and the fine print is the real story\x3C/p>\x3Cp style=\"margin:0 0 5px 0; font-size:15px; line-height:1.5; color:#0A0A0A;\">» The 20-minute page that decides whether any agent works for you\x3C/p>\x3Cp style=\"margin:0; font-size:15px; line-height:1.5; color:#0A0A0A;\">» Five subscriptions you can probably cancel this month\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">💸 The price broke\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"margin:16px 0 8px 0; font-size:19px; font-weight:700; line-height:1.3; color:#0A0A0A;\">Gemini Spark just got 80% cheaper\x3C/p>\x3Cp>Google added Spark to the $19.99 AI Pro plan in the US. It launched in May as an Ultra-only feature starting at $99.99 a month, so the cheapest way in just dropped by 80%.\x3C/p>\x3Ctable class=\"kit-stack\" role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 18px 0;\">\x3Ctr>\x3Ctd width=\"33%\" valign=\"top\" style=\"background-color:#F6F6F4; padding:12px 14px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#5F5E5A;\">Was\x3C/p>\x3Cp style=\"margin:2px 0 0 0; font-size:20px; font-weight:700; color:#5F5E5A;\">$99.99\x3C/p>\x3C/td>\x3Ctd width=\"8\">&nbsp;\x3C/td>\x3Ctd width=\"33%\" valign=\"top\" style=\"background-color:#E6F1FB; padding:12px 14px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#0074CA;\">Now\x3C/p>\x3Cp style=\"margin:2px 0 0 0; font-size:20px; font-weight:700; color:#0074CA;\">$19.99\x3C/p>\x3C/td>\x3Ctd width=\"8\">&nbsp;\x3C/td>\x3Ctd width=\"33%\" valign=\"top\" style=\"background-color:#F6F6F4; padding:12px 14px;\">\x3Cp style=\"margin:0; font-size:11px; letter-spacing:1px; text-transform:uppercase; color:#5F5E5A;\">Drop\x3C/p>\x3Cp style=\"margin:2px 0 0 0; font-size:20px; font-weight:700; color:#0A0A0A;\">80%\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Cp>What it actually is: three pieces. \x3Cstrong>Tasks\x3C/strong>, where you hand it something multi-step and walk away. \x3Cstrong>Schedules\x3C/strong>, where you set it to fire on a clock or a trigger. And \x3Cstrong>Skills\x3C/strong>, where you teach it how you do a thing once so you stop re-explaining it every time. It works across Gmail, Calendar, Docs, Sheets, Slides and Drive, and it keeps running when your laptop is shut.\x3C/p>\x3Cp>The examples Google leads with are the boring, correct ones. Scan the inbox every morning and summarize what matters. Pull meeting notes into a formatted report. Draft a reply when a specific person emails. Build a cited digest on a topic you follow.\x3C/p>\x3Cp>The piece worth your attention is Skills, because it's the closest thing anyone has shipped to a written job description for an agent. Google's own example: read the last 50 emails I wrote, turn that into a style guide for how I write, save it as a skill called ghostwriter, and call it every time I ask for a draft. That isn't a prompt. That's a documented process the agent reloads on every run. Hold that thought, because it's the whole second half of this email.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:0 0 18px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #0A0A0A; padding:14px 16px;\">\x3Cp style=\"margin:0; font-size:11px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0A0A0A;\">The catch\x3C/p>\x3Cp style=\"margin:6px 0 0 0; font-size:15px; line-height:1.6; color:#0A0A0A;\">Spark requires a personal Google account, and work or school accounts are excluded. On the $20 Pro plan it's US only. You also need to be over 18 with Keep Activity switched on.\x3C/p>\x3C/td>\x3C/tr>\x3C/table>\x3Cp>
1So if your business runs on Google Workspace, the agent that automates email cannot touch the inbox where your actual work lives. You'd be pointing it at a personal Gmail. The cost barrier came down and the permissions barrier stayed exactly where it was.\x3C/p>\x3Cp>There's a second read here, and it's the one I'd act on. Google just put an always-on agent at the same price as a chat subscription. That isn't a discount, it's a category move: background work is now the default thing you're paying for, and every other lab has to answer it. Which makes the useful question this week not \"should I subscribe.\" It's \x3Cstrong>which of my recurring work is specified well enough to hand over\x3C/strong> - because once the agent can only see half your accounts, the constraint stops being the tool and starts being you.\x3C/p>\x3Cp>\x3Ca href=\"https://gemini.google.com\" target=\"_blank\" rel=\"noopener noreferrer\">Try it here →\x3C/a>\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">🛠 Do this before Monday\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"margin:16px 0 8px 0; font-size:19px; font-weight:700; line-height:1.3; color:#0A0A0A;\">Write the job description first\x3C/p>\x3Cp>Every agent I've handed real work to has failed for the same reason, and it was never the model. I knew the task cold, I'd done it four hundred times, and I had never once written it down. So I'd type half an instruction, get back something almost right, fix it by hand, and quietly decide the tool wasn't there yet.\x3C/p>\x3Cp>The tool was fine. The spec was missing.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:14px 0 18px 0;\">\x3Ctr>\x3Ctd style=\"background-color:#F6F6F4; border-left:4px solid #00BC7D; padding:16px 18px;\">\x3Cp style=\"margin:0; font-size:17px; font-weight:700; color:#0A0A0A;\">The agent job description\x3C/p>\x3Cp style=\"margin:4px 0 14px 0; font-size:13px; color:#5F5E5A;\">20 minutes &middot; no purchase required\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"30\" valign=\"top\" style=\"padding:0 0 10px 0;\">\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"22\" height=\"22\" align=\"center\" valign=\"middle\" style=\"background-color:#00BC7D; font-size:12px; font-weight:700; color:#0A0A0A; line-height:22px;\">1\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3Ctd valign=\"top\" style=\"padding:0 0 10px 0; font-size:15px; line-height:1.55; color:#0A0A0A;\">Pick one thing you do every week that you resent. Inbox triage, chasing invoices, the weekly report.\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd width=\"30\" valign=\"top\" style=\"padding:0 0 10px 0;\">\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"22\" height=\"22\" align=\"center\" valign=\"middle\" style=\"background-color:#00BC7D; font-size:12px; font-weight:700; color:#0A0A0A; line-height:22px;\">2\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3Ctd valign=\"top\" style=\"padding:0 0 10px 0; font-size:15px; line-height:1.55; color:#0A0A0A;\">Do it once, out loud, narrating every decision you make. Record it on your phone if that's easier than typing.\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd width=\"30\" valign=\"top\" style=\"padding:0 0 10px 0;\">\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"22\" height=\"22\" align=\"center\" valign=\"middle\" style=\"background-color:#00BC7D; font-size:12px; font-weight:700; color:#0A0A0A; line-height:22px;\">3\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3Ctd valign=\"top\" style=\"padding:0 0 10px 0; font-size:15px; line-height:1.55; color:#0A0A0A;\">Write down the decision rules, not the steps.\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd width=\"30\" valign=\"top\" style=\"padding:0 0 10px 0;\">\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"22\" height=\"22\" align=\"center\" valign=\"middle\" style=\"background-color:#00BC7D; font-size:12px; font-weight:700; color:#0A0A0A; line-height:22px;\">4\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3Ctd valign=\"top\" style=\"padding:0 0 10px 0; font-size:15px; line-height:1.55; color:#0A0A0A;\">Write the stop conditions. What must it never do without asking you first?\x3C/td>\x3C/tr>\x3Ctr>\x3Ctd width=\"30\" valign=\"top\">\x3Ctable role=\"presentation\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\">\x3Ctr>\x3Ctd width=\"22\" height=\"22\" align=\"center\" valign=\"middle\" style=\"background-color:#00BC7D; font-size:12px; font-weight:700; color:#0A0A0A; line-height:22px;\">5\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3Ctd valign=\"top\" style=\"font-size:15px; line-height:1.55; color:#0A0A0A;\">Hand that page to whichever agent you already pay for.\x3C/td>\x3C/tr>\x3C/table>\x3C/td>\x3C/tr>\x3C/table>\x3Cp>Two things make or break this.\x3C/p>\x3Cp>\x3Cstrong>Decision rules, not steps.\x3C/strong> \"Check email\" is a step, and it's useless - the agent already knows how to open an inbox. \"Archive it if it's a newsletter I haven't opened in a month, flag it if a client's name is in the subject line, leave everything else alone\" is a rule. Rules are the part that lives in your head and nowhere else, which is exactly why the agent can't guess them and exactly why writing them down is the whole job.\x3C/p>\x3Cp>\x3Cstrong>Stop conditions.\x3C/strong> Name what it must never do without checking with you. Send anything to a client. Spend money. Delete instead of archive. Reply to anyone outside a list you write out. This is the line between an agent you trust with real work and one you supervise so closely that you'd have been faster doing it yourself.\x3C/p>\x3Cp>Twenty minutes, and you walk away with one page. \x3Cstrong>That page is the asset, not the subscription.\x3C/strong> Tools change every six weeks - the story above is a tool changing price by 80% in ten weeks - but a written process outlives all of them, and it's portable. The same page works in Spark, in Cowork, in whatever ships next quarter. And if you do end up in Spark, Skills is literally a place to store it.\x3C/p>\x3Cp>That's the difference between an agent that saves you an hour a day and one you quietly turn off on day three.\x3C/p>\x3Ctable role=\"presentation\" width=\"100%\" cellpadding=\"0\" cellspacing=\"0\" border=\"0\" style=\"margin:34px 0 0 0;\">\x3Ctr>\x3Ctd style=\"border-bottom:2px solid #0074CA; padding-bottom:6px; font-size:12px; font-weight:700; letter-spacing:2px; text-transform:uppercase; color:#0074CA;\">📺 Go deeper\x3C/td>\x3C/tr>\x3C/table>\x3Cp style=\"margin:16px 0 2px 0; font-size:19px; font-weight:700; line-height:1.3; color:#0A0A0A;\">5 SaaS tools you can cancel now\x3C/p>\x3Cp style=\"margin:0 0 12px 0; font-size:13px; color:#5F5E5A;\">▶ Video &middot; 10:41 &middot; free in the AI Resource Vault\x3C/p>\x3Cp>Calendly and Notion AI are two of the five. Calendly runs $12 to $25 a month, Notion AI is a $10 add-on on top of what you already pay Notion, so that's roughly $22 to $35 a month between those two alone.\x3C/p>\x3Cp>It isn't a cancel-everything pitch. For each of the five the guide lays out what replaces it, the exact setup, and where the replacement falls down - Calendly's public booking link is a real thing you give up, and if you're a coach taking 30 inbound bookings a week you should keep paying for it. Direction, not dogma.\x3C/p>\x3Cp>The other three are in the video.\x3C/p>\x3Cp>\x3Ca href=\"https://www.skool.com/try-ai-academy/classroom/8cd10fa9?md=f018242f245540f0849c8e7f5a454ad9\" target=\"_blank\" rel=\"noopener noreferrer\">Watch it in the vault →\x3C/a>\x3C/p>\x3Chr>\x3Cp>Do the job description one before Monday. Twenty minutes, costs nothing, and worst case you end up with a written version of something you've been carrying in your head for years - which is worth having whether or not you ever hand it to an agent.\x3C/p>\x3Cp>\x3Cstrong>That's the Toolkit.\x3C/strong>\x3C/p>\x3Cp>See you Tuesday for the Debrief.\x3C/p>\x3Cp>From Drew and The AI Debrief team\x3C/p>\x3Chr>\x3Cp>\x3Cstrong>P.S.\x3C/strong> What's the one task you'd hand an agent tomorrow if you actually trusted it? Hit reply and tell us. We read every one, and the good answers turn into future issues.\x3C/p>\x3Cp>\x3Cstrong>P.P.S.\x3C/strong> New here, or skipped one? Every past edition is in the \x3Ca href=\"https://thedebrief.ai/archive\" target=\"_blank\" rel=\"noopener noreferrer\">archive\x3C/a>.\x3C/p>\x3C/div>\x3C/td>\x3C/tr>\x3C/tbody>\x3C/table>",edition:"toolkit",issueNo:18},$R[168]={slug:"the-agent-model-that-fits-on-one-gpu",title:"The agent model that fits on one GPU",subtitle:"Meta just put a genuinely capable agent model on hardware you already own, and the license is the part almost nobody is talking about.",date:"2026-08-11",thumbnail:"https://embed.filekitcdn.com/e/jeCKAoLJGmzsPEwMDzFkXA/61aza6D7kVZhLTXBYu7h1K",blocks:$R[169]=[],html:"\x3Ctable cellPadding=\"0\" cellSpacing=\"0\" style=\"width:100%;margin:0 auto\">\x3Ctbody>\x3Ctr>\x3Ctd>\x3Cp class=\"\">Let&#x27;s debrief. \x3C/p>\x3Cp class=\"\">Meta just put a genuinely capable agent model on hardware you already own, and the license is the part almost nobody is talking about.\x3C/p>\x3Chr/>\x3Ch3 class=\"\">📌 The big thing\x3C/h3>\x3Cp class=\"\">\x3Cstrong style=\"font-weight:bolder\">Meta open-sourced a 30B agent model that runs on one consumer GPU\x3C/strong>\x3C/p>\x3Cp class=\"\">Meta released \x3Ca href=\"https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model\" target=\"_blank\" rel=\"noopener noreferrer\">Muse Glimmer\x3C/a>
1 on Monday: a 30-billion-parameter agentic model shipped under Apache 2.0. A 30B model normally needs over 55 GB of memory at full precision. Meta compressed it to roughly 4-bit, which drops it under 20 GB and fits it inside a single 24 GB or 32 GB card. On an RTX 5090 it decodes at 233 tokens per second. On an M4 Max, 38. Weights and quantized builds are on Hugging Face today.\x3C/p>\x3Cp class=\"\">\x3Cstrong style=\"font-weight:bolder\">The take:\x3C/strong> the license is the story, not the benchmark. Llama always shipped with usage restrictions attached. Apache 2.0 does not. Pair a permissive license with a model small enough to run on your own machine and the marginal cost of an always-on agent goes to roughly zero. When an agent costs nothing per call, you stop designing around how many calls you can afford and start designing around what you actually want it to do. That is a different kind of product.\x3C/p>\x3Cp class=\"\">\x3Cstrong style=\"font-weight:bolder\">Why you care:\x3C/strong> if you shelved a background agent because the API bill scaled with usage, that constraint just moved. One caveat before you build on it. Meta&#x27;s own numbers put Glimmer ahead on tool orchestration and behind Qwen3.6-27B on computer-use and terminal work, and those numbers have not been independently replicated yet. Treat it as a tool-calling brain, not a click-the-screen agent.\x3C/p>\x3Chr/>\x3Ch3 class=\"\">⚡ Worth your time\x3C/h3>\x3Cp class=\"\">→ The EU AI Act deadline everyone reported on August 2 was not the one that actually landed. High-risk obligations got pushed to December 2027. What turned on instead was Article 50 transparency: if you ship a chatbot to EU users it has to identify itself as AI, and synthetic media has to be labeled. That part is live right now. \x3Ca href=\"https://www.dataprotectionreport.com/2026/07/the-eu-ai-act-when-does-it-become-enforceable-now/\" target=\"_blank\" rel=\"noopener noreferrer\">What changed →\x3C/a>​\x3C/p>\x3Cp class=\"\">→ OpenAI split its Daybreak security program into two tiers and put GPT-5.6-Cyber behind the vetted one. In testing it found two unknown bugs in Chrome&#x27;s V8 engine that chain into a sandbox bypass, since patched. Notice the shape of it: the capability is here, and the thing gating it is an application form. \x3Ca href=\"https://www.axios.com/2026/08/10/openai-gpt-astra-restrictions-safety-hacking-defenders\" target=\"_blank\" rel=\"noopener noreferrer\">The details →\x3C/a>​\x3C/p>\x3Cp class=\"\">→ Full-duplex voice quietly got good this week. NVIDIA put out an open speech-to-speech model with roughly 450 ms turn-taking and live tool calling. If you tried building a voice agent a year ago and gave up on the lag, it is worth another hour of your time.\x3C/p>\x3Chr/>\x3Ch3 class=\"\">🙅 Skip it\x3C/h3>\x3Cp class=\"\">The &quot;Google&#x27;s AI will call stores for you&quot; headlines going around again. It is real, but it was announced last November and is still finishing its US rollout. Nothing actually changed this week. If you sell anything locally it is worth preparing for eventually, just not today.\x3C/p>\x3Chr/>\x3Cp class=\"\">The tools and the people behind this issue live in AI Academy. It is free, and it is where I answer things I cannot fit in an email. \x3Ca href=\"https://www.skool.com/try-ai-academy/about\" target=\"_blank\" rel=\"noopener noreferrer\">Come hang out →\x3C/a>​\x3C/p>\x3Cp class=\"\">\x3Cstrong style=\"font-weight:bolder\">That&#x27;s the Debrief.\x3C/strong>\x3C/p>\x3Cp class=\"\">See you Friday for the Toolkit.\x3C/p>\x3Cp class=\"\">Talk soon,\x3C/p>\x3Cp class=\"\">Drew\x3C/p>\x3Chr/>\x3Cp class=\"\">\x3Cstrong style=\"font-weight:bolder\">P.S.\x3C/strong> What are you building right now? Hit reply and tell me. I read every one.\x3C/p>\x3C/td>\x3C/tr>\x3C/tbody>\x3C/table>",edition:"debrief",issueNo:null},$R[170]={slug:"anthropic-just-filed-to-go-public-openai-is-next",title:"Anthropic just filed to go public (OpenAI is next)",subtitle:"The trillion-dollar AI IPO race just started",date:"2026-06-04",thumbnail:"/archive-images/HJbOdlwXQAIy7cZ-68cd9b6a.jpg",blocks:$R[171]=[$R[172]={t:"img",c:"/archive-images/Screenshot_2026-03-26_at_10.04.45_PM-5872dae0.png"},$R[173]={t:"p",c:"Anthropic just filed to go public - and OpenAI is days behind. For the first time ever, you'll be able to own a piece of the labs building AI. Meanwhile, Microsoft told developers Windows isn't for people anymore, Google wants to release 32 million mosquitoes (you've got until tomorrow to object), and GitHub Copilot users watched their $29 bill balloon toward $750 overnight."},$R[174]={t:"p",c:"Wall Street is about to decide which AI companies are worth a trillion dollars. This week, run the same diligence on your own stack - I'm showing you the only 5 AI tools I actually pay for."},$R[175]={t:"p",c:"\x3Cstrong>Let's debrief.\x3C/strong>"},$R[176]={t:"p",c:"In today's AI Debrief:"},$R[177]={t:"ul",c:$R[178]=["📰 Anthropic filed for an IPO - the trillion-dollar race to go public is on","🪟 Microsoft Build: Windows is now an agent platform","⚡ GitHub Copilot's new billing sent devs into revolt","🐬 The $10M prize to talk to animals named its finalists","🦟 Google wants to release 32 million mosquitoes (comment closes tomorrow)","🔧 Tool of the Week: The only 5 AI tools I pay for in 2026","🎓 Skill of the Week: The AI certifications every resume needs","🏠 Inside The AI Academy: 3,000 members + a new perks vault"]},$R[179]={t:"img",c:"/archive-images/Screenshot_2026-05-15_at_7.40.38_PM-a1277b6f.png"},$R[180]={t:"h2",c:"\x3Cstrong>ANTHROPIC JUST FILED FOR AN IPO - AND OPENAI IS DAYS BEHIND\x3C/strong>"},$R[181]={t:"img",c:"/archive-images/Screenshot_2026-06-04_at_1.06.53_PM-edfb3731.png"},$R[182]={t:"p",c:"Image Source: \x3Ca href=\"https://foxbusiness.com\" rel=\"noopener\" target=\"_blank\">foxbusiness.com\x3C/a>"},$R[183]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>On Monday, \x3Ca href=\"https://techcrunch.com/2026/06/01/anthropic-files-to-go-public/\" rel=\"noopener\" target=\"_blank\">Anthropic confidentially filed a draft S-1 with the SEC\x3C/a> for a proposed initial public offering - the first formal step toward letting the public own a piece of the lab behind Claude."},$R[184]={t:"p",c:"The details: The timing is no accident. The filing landed less than a week after Anthropic closed a $65 billion Series H that pushed its valuation to $965 billion - surpassing OpenAI's $852 billion and making it the most valuable AI startup on earth. OpenAI reportedly filed its own confidential prospectus around May 22 and is targeting a September debut at a $1 trillion-plus valuation . And SpaceX goe
1s first: its roadshow is underway this week with a Nasdaq debut planned for next week at a targeted $1.75 trillion valuation. Anthropic's numbers explain the confidence - its annualized revenue run-rate recently crossed $47 billion , up from roughly $10 billion a year ago, driven by enterprise adoption of Claude for coding and agent workflows. A confidential filing doesn't lock in a date - it lets Anthropic refine its disclosures with the SEC and pull the trigger when markets cooperate."},$R[185]={t:"p",c:"Why it matters: This is the moment the AI boom stops being a private-markets story. Until now, every \"OpenAI is worth $X\" headline was VCs marking their own homework. An S-1 means real audited financials - revenue by segment, losses, burn - in public for the first time , and Wall Street gets to vote on whether the AI trade is real. One analyst framed it perfectly: 2026 either becomes the most consequential IPO cycle since the dot-com era, or the most expensive lesson in narrative-versus-fundamentals public markets have ever taught . If you're a builder, the takeaway isn't \"buy the stock\" - it's that the companies whose tools you rent are about to face quarterly earnings pressure. Expect prices to go up, free tiers to shrink, and the gap between tools-worth-paying-for and tools-that-are-a-tax to get very wide, very fast."},$R[186]={t:"img",c:"/archive-images/Screenshot_2026-05-20_at_11.22.37_AM-f223121e.png"},$R[187]={t:"p",c:"🪟 Microsoft just made Windows work for robots, not people. At Build 2026 on Tuesday, Satya Nadella's thesis was that AI agents are now first-class citizens in Windows - in the runtime, the tooling, and how software gets distributed. Microsoft shipped seven new in-house MAI models (reasoning, coding, image, voice, transcription) and, in the quietest-but-biggest move, its own Project Polaris model will replace OpenAI inside GitHub Copilot by August . Translation: even Microsoft is cutting the OpenAI cord right before OpenAI asks Wall Street for a trillion dollars."},$R[188]={t:"p",c:"⚡ \x3Cstrong>\x3Ca href=\"https://www.buildfastwithai.com/blogs/ai-news-today-june-1-2026\" rel=\"noopener\" target=\"_blank\">GitHub Copilot's new billing sent developers into revolt.\x3C/a>\x3C/strong> As of June 1, Copilot switched from flat-fee subscriptions to token-based billing across all plans. The backlash was instant - one developer reported their $29/month bill is on track to hit roughly $750 under the new model, since heavy reasoning sessions that used to be silently subsidized now get charged at cost. \"What a joke\" became the shorthand across Reddit and X. The lesson for everyone, not just coders: the era of cheap, unlimited AI is ending, and your tool stack is about to get a lot more expensive if you're not deliberate about it."},$R[189]={t:"p",c:"🐬 \x3Cstrong>\x3Ca href=\"https://www.cnn.com/2026/06/03/science/animal-interspecies-communication\" rel=\"noopener\" target=\"_blank\">The $10M prize to actually talk to animals named its 2026 finalists.\x3C/a>\x3C/strong> The Coller-Dolittle Challenge - a five-year contest using AI and large language models to crack interspecies communication - revealed this year's contenders, with projects decoding African striped mice, marmosets, cuttlefish, and dolphins. The grand prize (a $10M investment or $500K cash) goes to the first team that gets an animal to communicate back without realizing it's talking to a human. AI just turned \"talking to animals\" from a fairy tale into an active research race."},$R[190]={t:"p",c:"🦟 \x3Cstrong>\x3Ca href=\"https://www.fastcompany.com/91551722/why-google-alphabet-releasing-32-million-mosquitoes-in-california-florida-mosquito-debug-project\" rel=\"noopener\" target=\"_blank\">Google wants to release 32 million mosquitoes - and you have until tomorrow to object.\x3C/a>\x3C/strong> Through Alphabet's \"Debug\" project, Google is seeking EPA approval to release up to 32 million sterilized male mosquitoes across California and Florida over two years to crash disease-carrying populations. The males can't bite, and their offspring never hatch. The public comment window closes June 5. It's a reminder that \"Google ships software\" is an outdated mental model - the company is now debugging the physical world."},$R[191]={t:"p",c:"\x3Cstrong>Wall Street is about to put a trillion-dollar price tag on the labs - and the second they answer to shareholders, the tools you rent get pricier and the free tiers get thinner (Copilot users already felt it: $29 to $750 overnight). But you don't have to overpay to keep up. Run the same diligence Wall Street is running, on your own stack. Here are the only 5 AI tools I actually pay for in 2026 - the ones that earn their spot every single month.\x3C/strong>"},$R[192]={t:"img",c:"/archive-images/Screenshot_2026-05-15_at_7.40.21_PM-d8702b13.png"},$R[193]={t:"h2",c:"\x3Cstrong>THE ONLY 5 AI TOOLS I PAY FOR IN 2026\x3C/strong>"},$R[194]={t:"p",c:"Everyone's stack is bloated. I audit mine every quarter and the list keeps shrinking - not growing. Out of the dozens of AI tools I've tested this year, only five survive the cut and earn a recurring charge on my card. Everything else is a free tier, a one-off, or a \"nice demo, no thanks.\""},$R[195]={t:"p",c:"\x3Cstrong>Why this matters now:\x3C/strong> This week proved the squeeze is real - Copilot users got hit with surprise bills, and every platform is racing to make you rent more. The skill that protects your wallet isn't finding more tools. It's ruthless curation: one tool per job, and the discipline to cut the rest."},$R[196]={t:"p",c:"\x3Cstrong>The 3-question audit (run it on your own stack this weekend):\x3C/strong>"},$R[197]={t:"ol",c:$R[198]=["\x3Cstrong>Does it do one job better than anything free?\x3C/strong> If a free tier covers 80% of the value, the paid tool has to justify the other 20% every month - not just the month you signed up.","\x3Cstrong>Would I notice if it vanished tomorrow?\x3C/strong> If you'd shrug, cancel it. The five I keep would each break a real workflow if they disappeared.","\x3Cstrong>Is it a tool or a tax?\x3C/strong> A tool makes you money or saves you hours. A tax is a subscription you forgot you had. Most stacks are 60% tax."]},$R[199]={t:"p",c:"🎁 \x3Ca href=\"https://www.skool.com/try-ai-academy/new-video-the-only-5-ai-tools-i-pay-for-in-2026\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>See the exact 5 tools I pay for (and why) inside The AI Academy - free →\x3C/strong>\x3C/a>"},$R[200]={t:"p",c:"👇 \x3Cstrong>Watch the YouTube walkthrough below\x3C/strong>"},$R[201]={t:"img",c:"/archive-images/maxresdefault-9426dd6d.jpg"},$R[202]={t:"p",c:"\x3Cstrong>THE ONLY 5 AI TOOLS I PAY FOR IN 2026\x3C/strong>"},$R[203]={t:"p",c:"\x3Cstrong>Stuck applying this? 3,000+ readers inside The AI Academy are figuring it out together. \x3C/strong>\x3Ca href=\"http://skool.com/try-ai-academy/about\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>Join free →\x3C/strong>\x3C/a>"},$R[204]={t:"img",c:"/archive-images/Screenshot_2026-05-15_at_7.42.46_PM-5b8da7cb.png"},$R[205]={t:"h2",c:"\x3Cstrong>THE AI CERTIFICATIONS EVERY RESUME NEEDS IN 2026\x3C/strong>"},$R[206]={t:"p",c:"This week, learn how to turn \"I use AI\" into something a hiring manager can actually verify. I rounded up the free and low-cost AI certifications worth adding to your resume right now - the ones that signal real skill, not just a ChatGPT login."},$R[207]={t:"p",c:"\x3Cstrong>Most people list \"AI\" as a skill and stop there. A certification turns a vague claim into proof - and proof is what gets the interview.\x3C/strong>"},$R[208]={t:"p",c:"\x3Cstrong>The 5-step play:\x3C/strong>"},$R[209]={t:"ol",c:$R[210]=["\x3Cstrong>Start with the universal ones.\x3C/strong> A handful of free certs from the major labs and platforms apply to almost any role and take an afternoon, not a semester. Stack these first.","\x3Cstrong>Add one role-specific cert.\x3C/strong> Marketing, ops, finance, design - there's a focused AI credential for your lane. Pick the one that matches the job you want next, not the job you have.","\x3Cstrong>Put them where they get seen.\x3C/strong> LinkedIn \"Licenses &amp; Certifications\" section first, resume skills section second. Recruiters filter on these fields.","\x3Cstrong>Pair each cert with a one-line proof.\x3C/strong> \"Certified in X - used it to automate Y, saving Z hours.\" A cert plus a result beats a cert alone every time.","\x3Cstrong>Refresh quarterly.\x3C/strong> AI certs date fast. Re-up the ones that still matter, drop the ones that don't."]},$R[211]={t:"p",c:"📘 \x3Cstrong>\x3Ca href=\"https://www.skool.com/try-ai-academy/new-video-20-free-ai-certifications-every-resume-needs-2\" rel=\"noopener\" target=\"_blank\">Grab the full certification list (with direct links) inside The AI Academy - free →\x3C/a>\x3C/strong>"},$R[212]={t:"p",c:"👇 \x3Cstrong>Watch the full walkthrough on YouTube\x3C/strong>"},$R[213]={t:"img",c:"/archive-images/maxresdefault-bfa4a3d2.jpg"},$R[214]={t:"p",c:"\x3Ca href=\"https://www.youtube.com/watch?v=9xFEfBFIJfY\" rel=\"noopener\" target=\"_blank\">\x3C/a> 20 Free AI Certifications Recruiters Actually Look For in 2026"},$R[215]={t:"p",c:"\x3Cstrong>Reading is the warm-up. Building is the workout. Join 3,000+ readers inside \x3C/strong>\x3Ca href=\"http://skool.com/try-ai-academy/about\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>The AI Academy →\x3C/strong>\x3C/a>"},$R[216]={t:"img",c:"/archive-images/Screenshot_2026-05-15_at_7.43.13_PM-7d4bb016.png"},$R[217]={t:"h2",c:"\x3Cstrong>INSIDE THE AI ACADEMY\x3C/strong>"},$R[218]={t:"p",c:"Two big moments this week. First, we crossed \x3Cstrong>3,000 members\x3C/strong> - the free community is officially one of the fastest-growing AI learning hubs anywhere, and it's still 100% free to join."},$R[219]={t:"p",c:"Second, we just opened a brand-new \x3Cstrong>Member Perks vault\x3C/strong> - exclusive discounts, free trials, and deals on the AI tools you actually use. Fitting, given this week's tool list: I pay full price for my five, but members are getting trials and discounts to test-drive them first. It's the easiest way to build your stack without the surprise bills the rest of the internet got this week."},$R[220]={t:"p",c:"🎁 \x3Cstrong>\x3Ca href=\"https://www.skool.com/try-ai-academy/classroom/97a007e1?md=7629e8631fa14d25935fa49feaccad39\" rel=\"noopener\" target=\"_blank\">See this month's member perks (free to claim) →\x3C/a>\x3C/strong>"},$R[221]={t:"img",c:"/archive-images/Screenshot_2026-05-20_at_11.30.22_AM-bb904e89.png"},$R[222]={t:"h2",c:"\x3Cstrong>EVERYTHING IN ONE PLACE\x3C/strong>"},$R[223]={t:"p",c:"🧰 \x3Cstrong>\x3Ca href=\"https://www.skool.com/try-ai-academy?s=newest-cm\" rel=\"noopener\" target=\"_blank\">The AI Academy \x3C/a>\x3C/strong>- our free AI community with over 1,000 members"},$R[224]={t:"p",c:"📚 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/archive\" rel=\"noopener\" target=\"_blank\">Archive\x3C/a>\x3C/strong> - every past edition in one place"},$R[225]={t:"p",c:"✍️ \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/write-for-us\" rel=\"noopener\" target=\"_blank\">Write for Us\x3C/a>\x3C/strong> - share your AI expertise with 75,000+ readers"},$R[226]={t:"p",c:"💸 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/referrals\" rel=\"noopener\" target=\"_blank\">Referrals\x3C/a>\x3C/strong> - refer friends, earn real rewards"},$R[227]={t:"p",c:"📣 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/advertise\" rel=\"noopener\" target=\"_blank\">Advertise\x3C/a>\x3C/strong> - reach 75,000+ AI professionals"},$R[228]={t:"p",c:"📣 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/contact-us\" rel=\"noopener\" target=\"_blank\">Contact Us \x3C/a>\x3C/strong>- get in touch with us for anything"},$R[229]={t:"img",c:"/archive-images/Screenshot_2026-05-15_at_7.44.04_PM-be89e21a.png"},$R[230]={t:"h2",c:"\x3Cstrong>Share the AI Debrief. Get paid. \x3C/strong>"},$R[231]={t:"p",c:"\x3Cstrong>Refer 3 →\x3C/strong> your name shouted out to 100,000 readers"},$R[232]={t:"p",c:"\x3Cstrong>Refer 10 →\x3C/strong> 50% off \x3Ca href=\"https://www.skool.com/ai-academy-plus/about\" rel=\"noopener\" target=\"_blank\">AI Academy Plus \x3C/a>"},$R[233]={t:"p",c:"\x3Cstrong>Refer 25 →\x3C/strong> Free month ($97) of \x3Ca href=\"https://www.skool.com/ai-academy-plus/about\" rel=\"noopener\" target=\"_blank\">AI Academy Plus \x3C/a>"},$R[234]={t:"p",c:"\x3Cstrong>This week's shoutouts:\x3C/strong>"},$R[235]={t:"p",c:"Want to see your name here next week? Refer one friend with your unique link below 👇"},$R[236]={t:"img",c:"/archive-images/Screenshot_2026-06-03_at_3.00.40_PM-dc7b99b6.png"},$R[237]={t:"h2",c:"🧰\x3Cstrong> DREW'S STACK\x3C/strong>"}
1,$R[238]={t:"p",c:"The tools I actually run The AI Debrief on. These are affiliate links, so if you sign up through them I earn a little at no extra cost to you. It genuinely helps keep this newsletter free and high-value, and I only list what I personally use."},$R[239]={t:"p",c:"\x3Cstrong>\x3Ca href=\"https://www.beehiiv.com/?via=Drew-Collins\" rel=\"noopener\" target=\"_blank\">Beehiiv - 20% off + free trial.\x3C/a>\x3C/strong> The newsletter platform powering this exact 100K-subscriber list."},$R[240]={t:"p",c:"\x3Cstrong>\x3Ca href=\"https://www.skool.com/signup?ref=1ce61877dca9482fbf418afc0ef69901\" rel=\"noopener\" target=\"_blank\">Skool - free trial.\x3C/a>\x3C/strong> The platform our whole community runs on, including the page 3,000 of you are already inside."},$R[241]={t:"p",c:"\x3Cstrong>\x3Ca href=\"https://www.gohighlevel.com/?fp_ref=founder-drew57\" rel=\"noopener\" target=\"_blank\">GoHighLevel - 30-day extended trial.\x3C/a>\x3C/strong> The all-in-one that replaces your CRM, funnels, email/SMS, and calendars in one login."},$R[242]={t:"p",c:"Clicking and signing up helps more than you'd think. Thanks for having my back."},$R[243]={t:"h2",c:"\x3Cstrong>Next Friday…\x3C/strong>"},$R[244]={t:"p",c:"WWDC is done and Apple finally showed its hand. I'll break down the new Siri, what's real vs. demo-ware, and the free alternative most people are sleeping on - so you can decide whether the \"new\" Apple AI is worth waiting for."},$R[245]={t:"p",c:"- Drew & the rest of the humans behind The AI Debrief"}],edition:"toolkit",issueNo:null},$R[246]={slug:"the-pope-just-picked-a-side-in-the-ai-war",title:"The Pope just picked a side in the AI war",subtitle:"42,000 words, one stark warning, and a $900B side note. Plus: build your own AI content team this weekend.",date:"2026-05-28",thumbnail:"/archive-images/el6oj254_pope-on-ai_625x300_26_May_26-a05d2abd.webp",blocks:$R[247]=[$R[248]={t:"img",c:"/archive-images/Screenshot_2026-03-26_at_10.04.45_PM-5872dae0.png"}
1,$R[249]={t:"p",c:"The Pope just sat next to an Anthropic co-founder, wrote a 42,000-word warning about AI, and called for the whole industry to slow down. Meanwhile, Anthropic is closing a $900 billion round - bigger than OpenAI. OpenAI's reasoning model just disproved an 80-year-old math problem on its own. And I built a 7-agent AI content team in Claude Code that runs my entire YouTube pipeline."},$R[250]={t:"p",c:"This week you can build one too."},$R[251]={t:"p",c:"Let's debrief."},$R[252]={t:"p",c:"In today's AI Debrief:"},$R[253]={t:"ul",c:$R[254]=["📰 The Pope's first encyclical: AI on trial at the Vatican","💰 Anthropic closing a $900B round (now bigger than OpenAI)","🧮 OpenAI's model solved an 80-year-old math problem solo","⚡ Cursor Composer 2.5: Opus-level coding at 1/10th the cost","🔧 Tool of the Week: Build your own AI content team in Claude Code","🎓 Skill of the Week: How I built a 50K newsletter ($3K/month)","🏠 Inside Try AI Academy: Jarvis, Apple giveaways, members helping members"]},$R[255]={t:"img",c:"/archive-images/Screenshot_2026-05-15_at_7.40.38_PM-a1277b6f.png"},$R[256]={t:"h2",c:"\x3Cstrong>POPE LEO PUT AI ON TRIAL - AND BROUGHT ANTHROPIC WITH HIM\x3C/strong>"},$R[257]={t:"img",c:"/archive-images/20260525T0700-BUDGET_NAME-1820289_web-63d132fe.jpg"},$R[258]={t:"p",c:"Image Source: \x3Ca href=\"https://www.ncronline.org/\" rel=\"noopener\" target=\"_blank\">ncronline.org\x3C/a>"},$R[259]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>On Monday, \x3Ca href=\"https://www.washingtonpost.com/technology/2026/05/25/anthropic-aligns-with-vatican-over-white-house-pope-leo-stokes-ai-fears/\" rel=\"noopener\" target=\"_blank\">Pope Leo XIV released his first encyclical\x3C/a> - a 42,000-word document called \"Magnifica Humanitas\" - calling for the \"disarming\" of artificial intelligence. Sitting next to him on stage: Anthropic co-founder Chris Olah."},$R[260]={t:"p",c:"The details: This is the first time a sitting pope has personally presented an encyclical to the world, and the first major Catholic teaching document specifically about AI. Leo signed it on May 15 - 135 years to the day after Leo XIII signed Rerum Novarum , the Industrial Revolution doctrine on workers' rights. The pope warned against the \"race for ever more powerful algorithms and larger datasets, driven by the desire to secure geopolitical or commercial dominance,\" and urged governments to slow AI development. Olah didn't play defense - he conceded that every frontier AI lab \"operates inside a set of incentives and constraints that can sometimes conflict with doing the right thing,\" and called on religious communities, civil society, and governments to push the technology toward better outcomes. The optics weren't subtle: Anthropic is already suing the Trump administration over its refusal to allow the U.S. military unrestricted use of Claude, and Leo's encyclical doubles down on monitoring AI in warfare."},$R[261]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>The most powerful religious institution in the world just framed AI as the defining moral test of our generation - and aligned with the lab that's positioning itself as the ethics-first alternative to OpenAI. For builders, this isn't background noise. It's the start of a regulatory and reputational divide: labs that lean into safety and constraints will keep winning institutional trust (and government contracts where ethics matter), while labs racing to ship will catch the headwinds. If you're building on top of these models, factor the politics into your stack now - the model you pick is a brand statement, not just a performance one."},$R[262]={t:"img",c:"/archive-images/Screenshot_2026-05-20_at_11.22.37_AM-f223121e.png"},$R[263]={t:"p",c:"💰 \x3Cstrong>\x3Ca href=\"https://finance.yahoo.com/sectors/technology/articles/anthropic-set-close-30-billion-203545596.html\" rel=\"noopener\" target=\"_blank\">Anthropic is closing a $900 billion round\x3C/a>\x3C/strong> - making it the most valuable AI startup on earth. Bloomberg reports Anthropic is wrapping a $30B+ raise this week led by Sequoia, Dragoneer, Altimeter, and Greenoaks. The valuation jump is brutal: $380B in February → $900B+ now. Anthropic is projecting Q2 revenue of $10.9B and a $50B annualized run rate by end of June. Three months ago OpenAI ($852B) was the king of AI valuations. This week, Anthropic took the crown."},$R[264]={t:"p",c:"🧮 OpenAI's reasoning model solved an 80-year-old math problem on its own. An unreleased general-purpose reasoning model disproved the Erdős unit distance conjecture - a discrete geometry problem first posed in 1946 - by producing a 125-page proof from a single open-ended prompt. Nine independent mathematicians, including Fields Medalist Tim Gowers, verified it . OpenAI is calling it \"the first time AI has autonomously solved a prominent open problem central to a field of mathematics.\" Translation: AI just stopped being a tool for math and started being a contributor to it."},$R[265]={t:"p",c:"⚡ \x3Cstrong>\x3Ca href=\"https://thenewstack.io/cursor-composer-benchmarks/\" rel=\"noopener\" target=\"_blank\">Cursor dropped Composer 2.5\x3C/a>\x3C/strong> - a new coding model that matches Claude Opus 4.7 and GPT-5.5 on benchmarks at roughly 1/10th the cost. It scores 79.8% on SWE-Bench Multilingual (Opus is 80.5%) and costs about $0.50 per task vs. $7 per task for Opus. It's only available inside Cursor, but the signal is bigger than the product: every IDE that doesn't ship its own model is about to feel the squeeze."},$R[266]={t:"p",c:"🍎 \x3Cstrong>\x3Ca href=\"https://www.bloomberg.com/news/features/2026-05-28/apple-ios-27-photos-screenshots-revamped-siri-pro-camera-app-new-ai-features\" rel=\"noopener\" target=\"_blank\">Apple's Siri overhaul is finally real.\x3C/a>\x3C/strong> Bloomberg got the first leaked screenshots of iOS 27 today, with a fully redesigned Siri and a new chatbot-style app as the centerpiece. Apple announces officially at WWDC on June 8. After 18 months of \"Apple is behind on AI\" takes, Cupertino is about to make its first real swing - and 1 billion+ iPhones will be the test bed."},$R[267]={t:"p",c:"\x3Cstrong>Big AI just consolidated more power this week - and even the Pope called it out. But you don't have to wait for the giants to share the leverage. Build your own AI team in Claude Code this weekend: seven agents that find trending topics, write scripts, design thumbnails, and run your entire content pipeline while you sleep.\x3C/strong>"},$R[268]={t:"img",c:"/archive-images/Screenshot_2026-05-15_at_7.40.21_PM-d8702b13.png"},$R[269]={t:"h2",c:"\x3Cstrong>BUILD YOUR OWN AI CONTENT TEAM IN CLAUDE CODE\x3C/strong>"},$R[270]={t:"p",c:"I just built a 7-agent system inside Claude Code that runs my YouTube pipeline end-to-e
1nd. Channel analyst studies my voice. Trend scout finds outlier videos every morning. Script writer drafts hooks in my tone. Thumbnail designer generates 3 concepts with my face composited in. Video editor cuts the raw recording. Daily reporter emails me a brief with everything ready to record."},$R[271]={t:"p",c:"\x3Cstrong>Why this matters now:\x3C/strong> Anthropic just hit $900B. The big labs are racing to build agents you'll rent for $100/month. Claude Code lets you build your own today - one project, seven specialized sub-agents, sharing state through Google Sheets. The ROI math is brutal: this stack costs ~$75-85/month and replaces a freelance social media manager, editor, and researcher who'd charge $4-7K combined."},$R[272]={t:"p",c:"\x3Cstrong>The 3-step start:\x3C/strong>"},$R[273]={t:"ol",c:$R[274]=["Install Claude Code ( npm install -g @anthropic-ai/claude-code ), grab your Anthropic API key, and create a single project folder with a CLAUDE.md file - this is the brain every agent reads first","Build the first loop only: Channel Analyst → Trend Scout → Script Writer → Daily Reporter. Don't try to build all 7 on day one. Get the idea-to-script-to-email loop working first","Layer in thumbnails, video editing, and shorts repurposing in week 2 once the core loop is humming"]},$R[275]={t:"p",c:"🎁\x3Cstrong>\x3Ca href=\"https://www.skool.com/try-ai-academy/classroom/8cd10fa9?md=3c22e266db96401dad5cd7123d30867e\" rel=\"noopener\" target=\"_blank\"> Grab the full step-by-step build guide inside The AI Academy (free) →\x3C/a>\x3C/strong>"},$R[276]={t:"p",c:"👇 \x3Cstrong>Watch the YouTube walkthrough below\x3C/strong>"},$R[277]={t:"img",c:"/archive-images/maxresdefault-b8e6228a.jpg"},$R[278]={t:"p",c:"\x3Ca href=\"https://www.youtube.com/watch?v=tSvLwVVHtmI\" rel=\"noopener\" target=\"_blank\">\x3C/a> 7 AI Agents Now Run My Entire YouTube Channel"},$R[279]={t:"p",c:"\x3Cstrong>Stuck applying this? 2,000+ readers inside The AI Academy are figuring it out together. \x3C/strong>\x3Ca href=\"http://skool.com/try-ai-academy/about\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>Join free →\x3C/strong>\x3C/a>\x3Cstrong> \x3C/strong>"},$R[280]={t:"img",c:"/archive-images/Screenshot_2026-05-15_at_7.42.46_PM-5b8da7cb.png"},$R[281]={t:"h2",c:"\x3Cstrong>HOW I BUILT A 50,000 AI NEWSLETTER ($3,000/MO)\x3C/strong>"},$R[282]={t:"p",c:"This week, learn the exact 3-layer stack that took The AI Debrief from zero to 50,000+ subscribers in 60 days - generating $3K/month while I sleep. No team. No ad budget beyond the engine itself. Just three tools wired together correctly."},$R[283]={t:"p",c:"\x3Cstrong>Most newsletters die at 1,000 subs because creators only build Layer 3 (writing). Build Layer 1 and 2 first.\x3C/strong>"},$R[284]={t:"p",c:"\x3Cstrong>The 5-step build:\x3C/strong>"},$R[285]={t:"ol",c:$R[286]=["\x3Cstrong>Layer 1 - Distribution:\x3C/strong> Turn on Beehiiv Boosts. Pay $1.50-$3.50 per confirmed subscriber. Start at $50/day, scale as ROI hits. This is the #1 growth lever and most creators skip it for \"real\" subscribers - Boosts subscribers ARE real","\x3Cstrong>Layer 2 - Conversion:\x3C/strong> Build a ManyChat IG DM funnel. Post a video, ask viewers to comment a keyword, auto-DM them the signup link. Beats a bio link 5-10x because the platform rewards comments and the DM feels personal","\x3Cstrong>Layer 3 - Content Engine:\x3C/strong> Build a Claude Project loaded with your prior issues, audience profile, and brand voice. Spend 4 hours upfront training it - this is the part everyone skips. Output: drafts that need 10 minutes of editing, not 2 hours of writing","\x3Cstrong>Pick a tighter niche than feels comfortable.\x3C/strong> \"AI\" is too broad. \"AI for marketers\" or \"AI for VCs\" converts better on Boosts, opens higher, and attracts better sponsors","\x3Cstrong>Monetize on the floor first:\x3C/strong> Beehiiv ad network + Boosts on the receiving end = ~$2K passive baseline. Direct sponsors are the variable upside. Don't chase sponsors until you have 10K+ engaged subs and a media kit"]},$R[287]={t:"p",c:"📘\x3Cstrong>\x3Ca href=\"https://www.skool.com/try-ai-academy/classroom/8cd10fa9?md=2b63b422a7d542538964572a9ca7e562\" rel=\"noopener\" target=\"_blank\"> \x3C/a>\x3C/strong>\x3Cstrong>\x3Ca href=\"https://www.skool.com/try-ai-academy/classroom/8cd10fa9?md=2b63b422a7d542538964572a9ca7e562\" rel=\"noopener\" target=\"_blank\">Grab the full playbook (Claude Project setup, ManyChat templates, Beehiiv Boosts categories) inside Try AI Academy (free) →\x3C/a>\x3C/strong>"},$R[288]={t:"p",c:"\x3Cstrong>👇 Watch the full walkthrough below\x3C/strong>"},$R[289]={t:"img",c:"/archive-images/maxresdefault-a4b37ebf.jpg"},$R[290]={t:"p",c:"\x3Ca href=\"https://www.youtube.com/watch?v=tkbxJtlXEPk\" rel=\"noopener\" target=\"_blank\">\x3C/a>
1 How I Built a 50K AI Newsletter in 60 Days (Now $3K/Mo)"},$R[291]={t:"p",c:"\x3Cstrong>Reading is the warm-up. Building is the workout. Join 2,000+ readers inside \x3C/strong>\x3Ca href=\"http://skool.com/try-ai-academy/about\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>The AI Academy →\x3C/strong>\x3C/a>"},$R[292]={t:"img",c:"/archive-images/Screenshot_2026-05-15_at_7.43.13_PM-7d4bb016.png"},$R[293]={t:"h2",c:"\x3Cstrong>INSIDE THE AI ACADEMY\x3C/strong>"},$R[294]={t:"p",c:"A great thread fired off this week: member Marwen B. asked \x3Ca href=\"https://www.skool.com/try-ai-academy/relationship-between-jarvis-and-claude\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>\"What's the relationship between Jarvis and Claude?\"\x3C/strong>\x3C/a> - kicking off a real conversation about local AI, APIs, and how to build your own assistant on top of Claude's brain. This is what's working inside the community: people walking through the guides together, breaking down what each tool actually does, and helping each other build instead of just consume."},$R[295]={t:"p",c:"Plus - \x3Cstrong>monthly Apple giveaways\x3C/strong> are still running. Active members get entered every month to win Apple gear (AirPods, iPads, MacBooks). Free to join, free to enter."},$R[296]={t:"p",c:"🍎\x3Cstrong>\x3Ca href=\"https://www.skool.com/try-ai-academy/monthly-apple-giveaways\" rel=\"noopener\" target=\"_blank\"> See this month's Apple giveaway →\x3C/a>\x3C/strong>"},$R[297]={t:"img",c:"/archive-images/Screenshot_2026-05-20_at_11.30.22_AM-bb904e89.png"},$R[298]={t:"h2",c:"\x3Cstrong>EVERYTHING IN ONE PLACE\x3C/strong>"},$R[299]={t:"p",c:"🧰 \x3Cstrong>\x3Ca href=\"https://www.skool.com/try-ai-academy?s=newest-cm\" rel=\"noopener\" target=\"_blank\">The AI Academy \x3C/a>\x3C/strong>- our free AI community with over 2,000 members"},$R[300]={t:"p",c:"📚 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/archive\" rel=\"noopener\" target=\"_blank\">Archive\x3C/a>\x3C/strong> - every past edition in one place"},$R[301]={t:"p",c:"✍️ \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/write-for-us\" rel=\"noopener\" target=\"_blank\">Write for Us\x3C/a>\x3C/strong> - share your AI expertise with 100,000+ readers"},$R[302]={t:"p",c:"💸 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/referrals\" rel=\"noopener\" target=\"_blank\">Referrals\x3C/a>\x3C/strong> - refer friends, earn real rewards"},$R[303]={t:"p",c:"📣 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/advertise\" rel=\"noopener\" target=\"_blank\">Advertise\x3C/a>\x3C/strong> - reach 100,000+ AI professionals"},$R[304]={t:"p",c:"📣 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/contact-us\" rel=\"noopener\" target=\"_blank\">Contact Us \x3C/a>\x3C/strong>- get in touch with us for anything"},$R[305]={t:"img",c:"/archive-images/Screenshot_2026-05-15_at_7.44.04_PM-be89e21a.png"},$R[306]={t:"h2",c:"\x3Cstrong>Share The AI Debrief. Get paid. \x3C/strong>"},$R[307]={t:"p",c:"\x3Cstrong>Refer 1 →\x3C/strong> your name shouted out to 100,000 readers"},$R[308]={t:"p",c:"\x3Cstrong>Refer 5 →\x3C/strong> mystery reward with real dollar value"},$R[309]={t:"p",c:"\x3Cstrong>Refer 10 →\x3C/strong> $100 back, every time. No cap. Forever."},$R[310]={t:"p",c:"\x3Cstrong>This week's shoutouts:\x3C/strong>"},$R[311]={t:"p",c:"Pieter V. · Saanvi R. · Tobias M. · Anuk D. · Felipe O. · Kenji H. · Mariella B. · Tomek K. · Yuki S. · Idris N. Pieter V. · Saanvi R. · Tobias M. · Anuk D. · Felipe O. · Kenji H. · Mariella B. · Tomek K. · Yuki S. · Idris N. · Caleb F. · Priya M. · Diego R. · Aanya V. · Ezekiel T. · Linnea J. · Rashid A. · Zaina H. · Soren P. · Naveen S. · Catalina M. · Bohdan Z. · Amaia L. · Theo G. · Wren E. · Olu B. · Mihail D. · Sienna O. · Yara K. · Kofi A. · Romina C. · Esteban N. · Anika T. · Ravi S. · Mateusz J. · Imani O. · Lucca F. · Nadya R. · Bilal A. · Iker S. · Camille D. · Olamide J. · Tomas R. · Cyrus M. · Avery K. · Roshni B. · Magnus E. · Selene V. · Tariq H. · Eliana P."},$R[312]={t:"p",c:"Want to see your name here next week? Refer one friend with your unique link below 👇"},$R[313]={t:"h2",c:"\x3Cstrong>NEXT FRIDAY…\x3C/strong>"},$R[314]={t:"p",c:"Apple announces the new Siri at WWDC on June 8. I'll break down what builders should actually pay attention to - and the free Apple alternative most people are sleeping on right now."},$R[315]={t:"p",c:"- Drew & the rest of the humans behind The AI Debrief"}],edition:"toolkit",issueNo:null},$R[316]={slug:"an-openai-founder-just-defected-to-anthropic",title:"An OpenAI founder just defected to Anthropic",subtitle:"Plus: Google's 24/7 AI agent, Meta cuts 8,000 jobs, and 10 image prompts I'm using this week.",date:"2026-05-20",thumbnail:"/archive-images/Screenshot_2026-05-20_at_11.18.10_AM-b4f760f6.png",blocks:$R[317]=[$R[318]={t:"img",c:"/archive-images/Screenshot_2026-03-26_at_10.04.45_PM-5872dae0.png"},$R[319]={t:"p",c:"One of OpenAI's original co-founders just jumped ship to Anthropic - and his job is to use Claude to build the next Claude. Google I/O dropped a 24/7 AI agent that runs while you sleep. Meta is cutting 8,000 jobs THIS week to fund a $145B AI war chest. And Nvidia reports tonight in what might be the most-watched earnings of the year."},$R[320]={t:"p",c:"I also dropped 10 image prompts that make content look like you hired a designer, and a 30-minute walkthrough to build your own AI chatbot for free."},$R[321]={t:"p",c:"Let's debrief."},$R[322]={t:"p",c:"\x3Cstrong>In today's AI Debrief:\x3C/strong>"},$R[323]={t:"ul",c:$R[324]=["📰 The AI defection that just rewrote the talent war","🤖 Google's 24/7 AI agent (it runs even when your laptop is closed)","💸 Meta cuts 8,000 jobs to fund its $145B AI bet","🎨 10 AI image prompts I'm using to crush content this week","🎓 Build your own AI chatbot in 30 minutes (free)","🏠 1,500 members + Apple giveaways inside the community"]},$R[325]={t:"img",c:"/archive-images/Screenshot_2026-05-15_at_7.40.38_PM-a1277b6f.png"},$R[326]={t:"h2",c:"\x3Cstrong>KARPATHY DEFECTS: ONE OF \x3C/strong>\x3Cstrong>\x3Ca href=\"https://techcrunch.com/2026/05/19/openai-co-founder-andrej-karpathy-joins-anthropics-pre-training-team/\" rel=\"noopener\" target=\"_blank\">OPENAI'S ORIGINAL CO-FOUNDERS JUST JOINED ANTHROPIC\x3C/a>\x3C/strong>"},$R[327]={t:"img",c:"/archive-images/Screenshot_2026-05-20_at_11.20.30_AM-84c521db.png"},$R[328]={t:"p",c:"Image Source: \x3Ca href=\"https://yourstory.com\" rel=\"noopener\" target=\"_blank\">yourstory.com\x3C/a>"}
1,$R[329]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>Andrej Karpathy, one of OpenAI's original co-founders and former Tesla AI lead, announced Tuesday that he's joined Anthropic. He started this week on the pre-training team under team lead Nick Joseph."},$R[330]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>This isn't a normal hire. Karpathy is going to lead a new group focused on using Claude to accelerate Claude's own pre-training research - basically, AI building AI. Pre-training is the most compute-intensive, most expensive phase of building a frontier model, so if Claude can meaningfully speed up its own training pipeline, the economics of the AI race shift fast. Anthropic's annualized revenue has tripled to $30 billion since December, and they've become a magnet for OpenAI talent: over the past two years OpenAI has lost more than a dozen senior executives and researchers, including former CTO Mira Murati and reinforcement learning pioneer John Schulman."},$R[331]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>Anthropic just signaled that the next moat isn't compute or data - it's whether your AI can train the next generation of AI. If recursive self-improvement actually works, the 12-18 month gap between model releases collapses. Translation for builders: the tools you're using today are going to feel obsolete way faster than they did last year. Don't wait for the perfect AI stack. Build with what works now and stay ready to swap."},$R[332]={t:"img",c:"/archive-images/Screenshot_2026-05-20_at_11.22.37_AM-f223121e.png"},$R[333]={t:"p",c:"🤖 Google I/O went full agent. Google unveiled Gemini Spark , a 24/7 personal AI agent that runs in the cloud and keeps working even when your laptop is closed - think monitoring credit card fraud, tracking school updates, or watching for specific deals. They also dropped Gemini 3.5 Flash with agentic capabilities, Gemini Omni for AI video generation, and bumped AI Ultra to $100/month with 5x the usage of AI Pro. The agent era has a price tag now. 💸 Meta starts cutting 8,000 jobs today. Meta begins eliminating roughly 10% of its workforce on Wednesday, May 20, leaving 6,000 open roles unfilled. The cuts come as Meta raises 2026 capex to $145 billion - up from $39.2 billion in 2024 - most of it Nvidia GPUs, AI infrastructure deals, and the Llama model ecosystem. The pattern is unmistakable: trim humans, scale compute. 📊 Nvidia reports tonight. Nvidia unveils its fiscal 2027 Q1 results after the close today, May 20. Analysts expect earnings of $1.78 per share, up 120% year over year, on revenue of $79.2 billion (up 79.5% YoY). Options pricing implies a ~6.5% move in the stock - a $350-355 billion swing in market value, larger than the market cap of 90% of S&P 500 companies. If you want one signal on whether AI capex is still accelerating, this is it. 🏢 OpenAI launched its own consulting army. OpenAI is launching the OpenAI Deployment Company, designed to embed Forward Deployed Engineers inside enterprises to build and deploy AI systems. It launches with more than $4 billion of initial investment and acquired Tomoro to bring in experienced FDEs from day one. OpenAI isn't just selling models anymore - they're selling implementation."},$R[334]={t:"p",c:"\x3Cstrong>OpenAI and Anthropic are racing to build the AI that builds AI. Cool. You don't need to wait for AGI to make AI work for you today - here are 10 image prompts that turn your phone into a content studio.\x3C/strong>"},$R[335]={t:"img",c:"/archive-images/Screenshot_2026-05-15_at_7.40.21_PM-d8702b13.png"},$R[336]={t:"h2",c:"\x3Cstrong>10 PLUG-AND-PLAY IMAGE PROMPTS\x3C/strong>"},$R[337]={t:"p",c:"10 prompts I'm using right now to crank out content that looks like I hired a designer. They work in ChatGPT, Claude, Midjourney, Gemini - whatever image model you're using."},$R[338]={t:"p",c:"\x3Cstrong>Why this matters now:\x3C/strong> Google's Gemini Omni for AI video is dropping soon. AI Ultra is $100/mo. The expensive tools are coming. But the prompts that turn AI image generation from \"meh, that looks like AI\" into \"wait, who designed this?\" - those work today, for free."},$R[339]={t:"p",c:"\x3Cstrong>How to use them:\x3C/strong>"},$R[340]={t:"ol",c:$R[341]=["Watch the 3-minute walkthrough below","Pick the prompt that matches what you need (LinkedIn carousel, YouTube thumbnail, hero image, social tile, etc.)","Swap in your variables (subject, mood, style)","Generate, tweak once, ship it"]},$R[342]={t:"p",c:"🎁\x3Cstrong>\x3Ca href=\"https://www.skool.com/try-ai-academy/classroom/8cd10fa9?md=2862cfb93cab4ac7854b5411aa173a82\" rel=\"noopener\" target=\"_blank\"> Grab the full prompt list inside Try AI Academy (free) →\x3C/a>\x3C/strong>"},$R[343]={t:"p",c:"👇 \x3Cstrong>Watch the YouTube walkthrough below\x3C/strong>"},$R[344]={t:"img",c:"/archive-images/maxresdefault-824bf333.jpg"},$R[345]={t:"p",c:"\x3Ca href=\"https://www.youtube.com/watch?v=LQp-P6iOxoM&amp;t=3s\" rel=\"noopener\" target=\"_blank\">\x3C/a>
1 10 AI Image Prompts That Don't Look Like AI"},$R[346]={t:"p",c:"\x3Cstrong>Stuck applying this? 1,500+ readers inside Try AI Academy are figuring it out together. \x3C/strong>\x3Ca href=\"http://skool.com/try-ai-academy/about\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>Join free →\x3C/strong>\x3C/a>\x3Cstrong> \x3C/strong>"},$R[347]={t:"img",c:"/archive-images/Screenshot_2026-05-15_at_7.42.46_PM-5b8da7cb.png"},$R[348]={t:"h2",c:"\x3Cstrong>BUILD YOUR OWN AI CHATBOT IN 30 MINUTES (FREE)\x3C/strong>"},$R[349]={t:"p",c:"This week, learn how to build a custom AI chatbot - no code required. Customer support bot, personal research assistant, knowledge base for your business, sales SDR that answers FAQs at 3am - you can spin one up in under 30 minutes, free."},$R[350]={t:"p",c:"\x3Cstrong>The 5-step build:\x3C/strong>"},$R[351]={t:"ol",c:$R[352]=["Pick your platform (Custom GPTs, Claude Projects, or a free open-source option - I'll show you which to choose)","Define the chatbot's one job in a single sentence (this is where 90% of people mess up)","Upload your knowledge base (PDFs, docs, FAQs, transcripts - whatever your bot needs to know)","Stress-test it with 5-10 real questions you'd expect users to ask","Embed it where people will actually use it - your site, Slack, Discord, or share the link"]},$R[353]={t:"p",c:"📘\x3Cstrong>\x3Ca href=\"https://www.skool.com/try-ai-academy/classroom/8cd10fa9?md=d11edaf4bfd3453599fc36a1e5ff283a\" rel=\"noopener\" target=\"_blank\"> Grab the step-by-step guide inside Try AI Academy →\x3C/a>\x3C/strong>"},$R[354]={t:"p",c:"👇 \x3Cstrong>Watch the full 30-minute tutorial below\x3C/strong>"},$R[355]={t:"img",c:"/archive-images/maxresdefault-d0035841.jpg"},$R[356]={t:"p",c:"\x3Ca href=\"https://www.youtube.com/watch?v=Lu8LlpPPOnA\" rel=\"noopener\" target=\"_blank\">\x3C/a> How To Build a Custom AI Chatbot in 30 Minutes (Free)"},$R[357]={t:"p",c:"\x3Cstrong>Reading is the warm-up. Building is the workout. Join 1,500+ readers inside \x3C/strong>\x3Cstrong>\x3Ca href=\"http://skool.com/try-ai-academy/about\" rel=\"noopener\" target=\"_blank\">Try AI Academy →\x3C/a>\x3C/strong>"},$R[358]={t:"img",c:"/archive-images/Screenshot_2026-05-15_at_7.43.13_PM-7d4bb016.png"},$R[359]={t:"h2",c:"\x3Cstrong>INSIDE THE AI ACADEMY\x3C/strong>"},$R[360]={t:"p",c:"We just crossed 1,500 members in the free community - and to celebrate, I dropped the monthly Apple giveaways . Active members get entered every month to win Apple gear (AirPods, iPads, MacBooks - the prize scales as the community grows). Free to join, free to enter."},$R[361]={t:"p",c:"🍎\x3Cstrong>\x3Ca href=\"https://www.skool.com/try-ai-academy/monthly-apple-giveaways\" rel=\"noopener\" target=\"_blank\"> See this month's Apple giveaway →\x3C/a>\x3C/strong>"},$R[362]={t:"img",c:"/archive-images/Screenshot_2026-05-20_at_11.30.22_AM-bb904e89.png"},$R[363]={t:"h2",c:"\x3Cstrong>EVERYTHING IN ONE PLACE\x3C/strong>"},$R[364]={t:"p",c:"🧰 \x3Cstrong>\x3Ca href=\"https://www.skool.com/try-ai-academy?s=newest-cm\" rel=\"noopener\" target=\"_blank\">The AI Academy \x3C/a>\x3C/strong>- our free AI community with over 1,000 members"},$R[365]={t:"p",c:"📚 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/archive\" rel=\"noopener\" target=\"_blank\">Archive\x3C/a>\x3C/strong> - every past edition in one place"},$R[366]={t:"p",c:"✍️ \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/write-for-us\" rel=\"noopener\" target=\"_blank\">Write for Us\x3C/a>\x3C/strong> - share your AI expertise with 75,000+ readers"},$R[367]={t:"p",c:"💸 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/referrals\" rel=\"noopener\" target=\"_blank\">Referrals\x3C/a>\x3C/strong> - refer friends, earn real rewards"},$R[368]={t:"p",c:"📣 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/advertise\" rel=\"noopener\" target=\"_blank\">Advertise\x3C/a>\x3C/strong> - reach 75,000+ AI professionals"},$R[369]={t:"p",c:"📣 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/contact-us\" rel=\"noopener\" target=\"_blank\">Contact Us \x3C/a>\x3C/strong>- get in touch with us for anything"},$R[370]={t:"img",c:"/archive-images/Screenshot_2026-05-15_at_7.44.04_PM-be89e21a.png"},$R[371]={t:"h2",c:"\x3Cstrong>Share the AI Debrief. Get paid. \x3C/strong>"},$R[372]={t:"p",c:"\x3Cstrong>Refer 1 →\x3C/strong> your name shouted out to 75,000 readers"},$R[373]={t:"p",c:"\x3Cstrong>Refer 5 →\x3C/strong> mystery reward with real dollar value"},$R[374]={t:"p",c:"\x3Cstrong>Refer 10 →\x3C/strong> $100 back, every time. No cap. Forever."},$R[375]={t:"p",c:"Big shoutout to this week's top referrers and most active members inside the community:"},$R[376]={t:"p",c:"\x3Cstrong>Hornyak P. · Adam C. · Ariana Y. · Numaan W. · Ryton O. · Gopu A. · Royi K. · Alex W. · Jayden K. · Sokratis L.\x3C/strong>"},$R[377]={t:"p",c:"Want to see your name here next week? Refer one friend with your unique link below 👇"},$R[378]={t:"h2",c:"\x3Cstrong>NEXT FRIDAY…\x3C/strong>"},$R[379]={t:"p",c:"The AI workflow that saved me 10+ hours last week. One free tool, three connected prompts, and a setup most people get wrong on step one. I'll walk through the exact build."},$R[380]={t:"p",c:"- Drew & the rest of the humans behind The AI Debrief"}],edition:"debrief",issueNo:null},$R[381]={slug:"the-fda-for-ai",title:"The FDA for AI",subtitle:"Trump wants to approve AI before you can use it.",date:"2026-05-07",thumbnail:"/archive-images/FDA_AI-f921b39c.webp",blocks:$R[382]=[$R[383]={t
1:"img",c:"/archive-images/Screenshot_2026-03-26_at_10.04.45_PM-5872dae0.png"},$R[384]={t:"p",c:"The White House just compared AI regulation to the FDA - and they're not joking. Google got caught sneaking a 4-gigabyte AI model onto your computer while you weren't looking. And the EU pulled an all-nighter to rewrite the rules that govern every AI company on Earth. Let's get into it."},$R[385]={t:"p",c:"\x3Cstrong>In today’s AI debrief:\x3C/strong>"},$R[386]={t:"ul",c:$R[387]=["The White House wants to vet AI models before they're released to the public","Google Chrome is silently installing a 4GB AI on your computer","The EU just rewrote its AI Act in an all-night negotiation"]},$R[388]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_5.51.07_PM-fab03707.png"},$R[389]={t:"h2",c:"\x3Cstrong>THE WHITE HOUSE WANTS AN \"FDA FOR AI\"\x3C/strong>"},$R[390]={t:"img",c:"/archive-images/0bf8fe42d9a2a66ebe618e5869cc172b-eae155de.webp"},$R[391]={t:"p",c:"Image Source: \x3Ca href=\"https://aol.com\" rel=\"noopener\" target=\"_blank\">aol.com\x3C/a>"},$R[392]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>The Trump administration is actively considering an executive order that would create a government review process for new AI models before they're released to the public - essentially an FDA-style approval system for artificial intelligence."},$R[393]={t:"p",c:"The details: Kevin Hassett, director of the National Economic Council, laid it out on Fox Business on Wednesday : \"We're studying possibly an executive order to give a clear road map to everybody about how this is going to go and how future AI that also potentially create vulnerabilities should go through a process so that they're released in the wild after they've been proven safe, just like an FDA drug.\" The comments came days after the New York Times reported that White House officials had already briefed executives from Anthropic, Google, and OpenAI on the plans during private meetings last week. The proposed executive order would create a government-industry \"AI working group\" bringing together tech executives, the NSA, the Office of the National Cyber Director, and the Director of National Intelligence. The catalyst is impossible to ignore: Anthropic's Claude Mythos Preview - a model so powerful it autonomously discovered thousands of zero-day vulnerabilities in every major operating system and web browser, some hiding in plain sight for over 25 years. Then the UK's AI Security Institute tested OpenAI's GPT-5.5 on the same gauntlet and found it scored 71.4% on expert-level cybersecurity tasks versus Mythos's 68.6% - essentially a tie. AISI's conclusion was blunt: these offensive capabilities aren't unique to one lab. They're an emergent byproduct of general improvements in reasoning, coding, and autonomy across all frontier models. A White House official told Reuters they wouldn't confirm or deny the report: \"Any policy announcement will come directly from the president.\""},$R[394]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>This is a 180-degree turn from January 2025, when Trump revoked Biden's AI safety executive order on his first day in office. The irony is hard to miss - the administration that dismantled the last round of AI guardrails is now proposing something potentially more aggressive. An FDA-style review process would fundamentally change the release cadence for every frontier lab. Open-source projects face the hardest enforcement questions. And the global implications are massive: the US would be moving closer to the regulatory posture the UK's AI Security Institute already operates, narrowing one of the biggest transatlantic policy gaps. For builders and investors, the message is clear - the era of \"ship it and see what happens\" with frontier models may be ending. The question is whether government review can move at the speed of AI development, or whether it becomes the bottleneck that pushes the next breakthrough overseas."},$R[395]={t:"h2",c:"\x3Cstrong>GOOGLE CHROME IS SECRETLY INSTALLING AI ON YOUR COMPUTER\x3C/strong>"},$R[396]={t:"img",c:"/archive-images/f8511b25-ce01-44d3-9385-8de6541c4db9-881e6578.jpeg"},$R[397]={t:"p",c:"Image Source: \x3Ca href=\"https://tech.yahoo.com\" rel=\"noopener\" target=\"_blank\">tech.yahoo.com\x3C/a>"},$R[398]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>Security researcher Alexander Hanff discovered that Google Chrome is silently downloading a 4GB AI model called Gemini Nano onto users' devices - without notification, without consent, and if you delete it, Chrome downloads it again."},$R[399]={t:"p",c:"The details: Hanff, a computer scientist and lawyer who publishes as \"That Privacy Guy,\" ran a controlled test using a fresh Chrome profile on macOS with zero human input. The browser evaluated the system's hardware, deemed it eligible, and downloaded the full 4GB payload during what appeared to be idle time - completing in just over fourteen minutes . The file is called \"weights.bin\" and sits in a folder labeled OptGuideOnDeviceModel. It's been getting bigger over time - 3GB in April 2025, 4GB by November 2025 . But here's the kicker Hanff flagged: Chrome 147 now renders an \"AI Mode\" pill in the address bar that looks like it's powered by the local model. It's not. Every query typed into AI Mode is sent to Google's cloud servers for processing. The local model sits on your disk unused while the cloud-backed version collects your data. At Chrome's global scale - roughly 3.5 billion devices - the mass distribution generates an estimated 6,000 to 60,000 tonnes of CO2-equivalent emissions . Hanff argues the behavior violates the EU's ePrivacy Directive and GDPR, and noted that Anthropic's Claude Desktop app pulled a similar move - quietly installing a browser integration bridge across multiple Chromium browsers without asking."},$R[400]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>This is what happens when the AI race hits distribution. Google doesn't need you to opt in to Gemini - it just needs you to not opt out, and most people won't because they don't know it's happening. The pattern is becoming familiar: ship AI features as defaults, bury the toggle, and count on user inertia to build the install base. If you're using Chrome right now, there's a good chance 4GB of your storage is occupied by an AI model you never asked for and can't permanently remove without disabling experimental flags. The broader signal for the industry is clear - the consent model around on-device AI is broken, and regulators in the EU are watching. For anyone building AI-powered tools, this is a case study in how not to deploy: even if the technology works, silent installation erodes the trust you need to keep users long-term."},$R[401]={t:"h2",c:"\x3Cstrong>THE EU JUST PULLED AN ALL-NIGHTER TO REWRITE THE AI ACT\x3C/strong>"},$R[402]={t:"img",c:"/archive-images/2023-12-09T000000Z_504550871_MT1NURPHO000HNJR28_RTRMADP_3_EU-d9cda140.jpg"},$R[403]={t:"p",c:"Image Source: \x3Ca href=\"https://atlanticcouncil.org\" rel=\"noopener\" target=\"_blank\">
1atlanticcouncil.org\x3C/a>"},$R[404]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>The EU Council and European Parliament struck a provisional deal at dawn on May 7 to simplify the AI Act - pushing back deadlines, extending SME exemptions, and adding a new ban on AI-generated deepfake pornography and child abuse material."},$R[405]={t:"p",c:"The details: The deal, reached after negotiations that ran through the night, is part of the EU's Omnibus VII simplification package . The biggest change: high-risk AI rules that were scheduled to take effect on August 2, 2026, are now delayed. Stand-alone high-risk systems get until December 2, 2027 . High-risk systems embedded in products like medical devices, machinery, and toys get until August 2, 2028. The watermarking deadline for AI-generated content transparency moved to December 2, 2026. SME exemptions now extend to small mid-cap enterprises. And the deal clarifies that the AI Office handles oversight where a single provider builds both the general-purpose model and the AI system on top of it - but national authorities keep jurisdiction over law enforcement, border management, courts, and financial institutions. The provision that broke new ground: a prohibition on AI systems that generate child sexual abuse material or create non-consensual intimate imagery of identifiable people. The European Commission proposed this package just five months ago and called it the fastest legislative turnaround on a major digital file in EU history."},$R[406]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>The AI Act was supposed to be done. Companies had been scrambling to meet the August 2, 2026 deadline. Now they have 16 extra months for stand-alone systems and 24 months for embedded ones - because the harmonized standards and conformity assessment tools simply weren't ready. For US companies operating in Europe, this is a reprieve, but not a pardon. The architecture of the law stands. The penalties still reach EUR 35 million or 7% of worldwide turnover. And the nudifier ban signals that Parliament will keep adding obligations to any legislative vehicle that moves. If you build AI products that touch Europe, the playbook hasn't changed - prepare for compliance, just on a slightly longer timeline. The real takeaway: even regulators are admitting they can't keep pace with AI development. The law designed to govern the most powerful technology of the century needed a rewrite before it even took effect."},$R[407]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.00.12_PM-ef59bfe3.png"},$R[408]={t:"ul",c:$R[409]=["\x3Cstrong>Anthropic is now the highest-revenue AI company in the world.\x3C/strong> Anthropic hit \x3Ca href=\"https://www.saastr.com/anthropic-just-passed-openai-in-revenue-while-spending-4x-less-to-train-their-models/\" rel=\"noopener\" target=\"_blank\">$30 billion in annualized recurring revenue\x3C/a> in April, passing OpenAI's $24 billion - while spending roughly 4x less on model training. The jump from $9B to $30B happened in about eight weeks. Over 1,000 companies now spend more than $1 million per year on Claude. OpenAI disputes the figure, arguing it overstates revenue by roughly $8 billion due to accounting differences. Both companies are targeting IPOs in the second half of 2026.","\x3Cstrong>Only 17.8% of the world's working-age population uses AI.\x3C/strong> Microsoft's new \x3Ca href=\"https://blogs.microsoft.com/on-the-issues/2026/05/07/the-state-of-global-ai-diffusion-in-2026/\" rel=\"noopener\" target=\"_blank\">Global AI Diffusion Report\x3C/a> shows AI adoption rose 1.5 percentage points in Q1 2026. The UAE leads at 70.1%. The US ranks 21st at 31.3%. The surprise: software developer employment hit a record 2.2 million in 2025, up 8.5% year over year - suggesting AI coding tools may be increasing demand for developers, not replacing them.","OpenAI shelved plans to spin out its robotics and hardware divisions before its IPO. The Wall Street Journal reported that OpenAI considered separating the units into independent entities but concluded they'd remain on its balance sheet regardless . The decision keeps OpenAI's IPO story simpler but signals the company's hardware ambitions are real enough to warrant a spinout discussion.","\x3Cstrong>
1IREN is acquiring cloud infrastructure firm Mirantis in an all-stock deal worth roughly $625 million.\x3C/strong> The deal adds \x3Ca href=\"https://www.cryptointegrat.com/p/ai-news-may-6-2026\" rel=\"noopener\" target=\"_blank\">Kubernetes orchestration and enterprise support\x3C/a> to IREN's AI cloud platform. The AI infrastructure build-out continues to accelerate as compute demand outstrips capacity.","Trump's White House is weighing government reviews of AI models before public release - and the move was prompted by Anthropic's Mythos model demonstrating offensive cybersecurity capabilities that matched or exceeded human experts. The UK's AI Security Institute found that GPT-5.5 hit similar levels , suggesting the risk is industry-wide."]},$R[410]={t:"img",c:"/archive-images/community_wins__1_-85006d73.png"},$R[411]={t:"h2",c:"\x3Cstrong>Share the AI Debrief. Get paid. \x3C/strong>"},$R[412]={t:"p",c:"\x3Cstrong>Refer 1 →\x3C/strong> your name shouted out to 25,000 readers"},$R[413]={t:"p",c:"\x3Cstrong>Refer 5 →\x3C/strong> mystery reward with real dollar value"},$R[414]={t:"p",c:"\x3Cstrong>Refer 10 →\x3C/strong> $100 back, every time. No cap. Forever."},$R[415]={t:"p",c:"Your referral link is below - one share is all it takes."},$R[416]={t:"p",c:"\x3Cstrong>Big thanks to these 151 readers who referred at least one person this week. You're the reason this newsletter keeps growing:\x3C/strong>"},$R[417]={t:"p",c:"Amara T. · Jensen K. · Priya S. · Beckett L. · Zuri M. · Cade R. · Nalini W. · Otis F. · Celeste D. · Ibrahim G. · Wren H. · Marcelo V. · Tessa J. · Kian P. · Rosalie B. · Dimitri N. · Sable E. · Arlo C. · Juniper A. · Raheem S. · Lyra K. · Camden W. · Bianca F. · Soren D. · Mila T. · Jasper H. · Aniyah R. · Callum L. · Daphne G. · Eshan M. · Frankie P. · Hadley B. · Idris C. · Jolene N. · Kenji V. · Leona S. · Matteo A. · Nia K. · Omar W. · Poppy J. · Quinn F. · Reina D. · Samir T. · Thalia H. · Ulysses R. · Vera L. · Winston G. · Xiomara M. · Yosef P. · Zola B. · Adrian E. · Brielle C. · Cyrus N. · Dahlia V. · Emmett S. · Freya A. · Gage K. · Harlow W. · Inara F. · Jude D. · Kamila T. · Lachlan H. · Maeve R. · Niko L. · Opal G. · Phoenix M. · Raya P. · Stellan B. · Tatum E. · Uma C. · Viggo N. · Waverly V. · Xander S. · Yasmin A. · Zeke K. · Alessia W. · Byron F. · Corinne D. · Declan T. · Elara H. · Finley R. · Gianna L. · Heath G. · Ilana M. · Joaquin P. · Kaia B. · Lennox E. · Miriam C. · Nash N. · Orla V. · Penn S. · Ramona A. · Silas K. · Tamsin W. · Uri F. · Valentina D. · Wells T. · Ximena H. · Yael R. · Zander L. · Amira G. · Bodhi M. · Cleo P. · Dashiell B. · Eloise E. · Fletcher C. · Gemma N. · Henrik V. · Isadora S. · Jules A. · Khalil K. · Liora W. · Magnus F. · Nola D. · Orion T. · Petra H. · Ronan R. · Sage L. · Theron G. · Vienna M. · Wilder P. · Zadie B. · Alistair E. · Belen C. · Cassian N. · Dove V. · Elio S. · Fiona A. · Greyson K. · Hana W. · Iver F. · Jaya D. · Kenzo T. · Lila H. · Mercer R. · Nadia L. · Oakley G. · Paloma M. · Rio P. · Sloane B. · Tobias E. · Vesper C. · Willa N. · Yannis V. · Zinnia S."},$R[418]={t:"p",c:"\x3Cstrong>Want your name here next week? One referral is all it takes.\x3C/strong>"},$R[419]={t:"p",c:"- Drew & the rest of the humans behind The AI Debrief"}],edition:"toolkit",issueNo:null},$R[420]={slug:"musk-wants-130-billion-back",title:"Musk Wants $130 Billion Back",subtitle:"OpenAI's on trial. China's drawing lines. And Wall Street's getting automated.",date:"2026-04-30",thumbnail:"/archive-images/4d747be3-227e-47bb-84ae-426b76ca06db_887x588-5dba5e9c.jpg",blocks:$R[421]=[$R[422]={t:"img",c:"/archive-images/Screenshot_2026-03-26_at_10.04.45_PM-5872dae0.png"},$R[423]={t:"p",c:"Elon Musk just took the stand and told a jury that OpenAI \"could also kill us all\" - while asking for $130 billion in damages. China killed Meta's biggest AI acquisition with a 54-character decree. And a 25-year-old Princeton grad just raised $160 million to replace your junior banker. Let's get into it."},$R[424]={t:"p",c:"\x3Cstrong>In today’s AI debrief:\x3C/strong>"},$R[425]={t:"ul",c:$R[426]=["Musk takes the stand in his $130B lawsuit against OpenAI","China blocks Meta's $2 billion Manus acquisition","Rogo raise
1s $160M to build AI agents for Wall Street"]},$R[427]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_5.51.07_PM-fab03707.png"},$R[428]={t:"h2",c:"\x3Cstrong>MUSK VS. OPENAI: $130 BILLION ON THE LINE\x3C/strong>"},$R[429]={t:"img",c:"/archive-images/ai-bubble-could-pop-within-days-if-musk-wins-the-lawsuit-on--21b21e40.webp"},$R[430]={t:"p",c:"Image Source: \x3Ca href=\"https://www.reddit.com/r/BetterOffline/comments/1szfflh/ai_bubble_could_pop_within_days_if_musk_wins_the/\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>Reddit\x3C/strong>\x3C/a>"},$R[431]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>Elon Musk spent three days on the witness stand this week in an Oakland federal courtroom, accusing OpenAI, Sam Altman, and Greg Brockman of betraying the nonprofit mission he helped create - and asking a jury to award $130 billion in damages."},$R[432]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>Musk co-founded OpenAI as a nonprofit in 2015, contributing what he says was \x3Ca href=\"https://www.cnn.com/2026/04/28/tech/elon-musk-sam-altman-openai\" rel=\"noopener\" target=\"_blank\">at least $44 million\x3C/a> in its early years. His core argument: he deliberately chose a nonprofit structure to build AI for the public good, and Altman and Brockman converted it into a for-profit to enrich themselves. OpenAI completed its restructuring in October and \x3Ca href=\"https://www.nbcnews.com/tech/tech-news/elon-musk-openai-sam-altman-trial-day-two-testimony-cross-examination-rcna342660\" rel=\"noopener\" target=\"_blank\">closed a $122 billion funding round\x3C/a> last month. Musk said he was fine with a for-profit subsidiary \"as long as it was a subsidiary of the nonprofit,\" but added: \x3Ca href=\"https://www.cnn.com/2026/04/29/tech/musk-openai-altman-lawsuit-testimony-day-two\" rel=\"noopener\" target=\"_blank\">\"What you can't have is the for-profit become the main event.\"\x3C/a> Cross-examination got heated fast. OpenAI's lead attorney Bill Savitt pressed Musk on whether he'd pushed for a for-profit structure himself, citing old emails. Musk fired back that the questions were \x3Ca href=\"https://www.nbcnews.com/tech/tech-news/elon-musk-openai-sam-altman-trial-day-two-testimony-cross-examination-rcna342660\" rel=\"noopener\" target=\"_blank\">\"designed to trick me.\"\x3C/a> When Savitt pointed out Musk hadn't actually donated $100 million despite claiming so, Musk called the questioning \"misleading.\" Savitt's opening statement framed it differently: \x3Ca href=\"https://www.cnn.com/2026/04/28/tech/elon-musk-sam-altman-openai\" rel=\"noopener\" target=\"_blank\">\"We are here because Mr. Musk turned out to be very wrong about OpenAI.\"\x3C/a> Greg Brockman has been given 48 hours' notice to testify next. Musk is seeking Altman's removal from the board, the company's return to nonprofit status, and $130 billion in damages going back to OpenAI's foundation."},$R[433]={t:"p",c:"Why it matters: This trial is happening as OpenAI plans what could be the largest tech IPO in history . A jury verdict favoring Musk could force a structural unwind that would vaporize billions in investor value and reshape who controls the most powerful AI company on Earth. But even if Musk loses, the testimony is putting OpenAI's origin story on public record - every email, every internal power struggle, every broken promise - right as they're trying to sell that story to public markets. For builders and investors watching the AI space, the real question isn't who wins the trial. It's whether the nonprofit-to-for-profit playbook that funded the entire frontier AI race just got permanently discredited."},$R[434]={t:"h2",c:"\x3Cstrong>CHINA KILLS META'S $2 BILLION MANUS DEAL\x3C/strong>"},$R[435]={t:"img",c:"/archive-images/0x0-1-fbcfef5a.webp"},$R[436]={t:"p",c:"Image Source: \x3Ca href=\"https://usaherald.com/china-moves-to-kill-metas-2-billion-ai-deal-sending-shockwaves-through-global-tech/\" rel=\"noopener\" target=\"_blank\">USA Herald \x3C/a>"},$R[437]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>China's top economic planner ordered Meta to unwind its $2 billion acquisition of Manus - the agentic AI startup that relocated from Beijing to Singapore before Zuckerberg scooped it up. The deal is effectively dead."},$R[438]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>Meta announced the Manus acquisition in December 2025, planning to fold its \x3Ca href=\"https://techcrunch.com/2026/04/27/china-vetoes-metas-2b-manus-deal-after-months-long-probe/\" rel=\"noopener\" target=\"_blank\">
1autonomous AI agent technology\x3C/a> into Meta AI. Manus was founded in China in 2022 and moved to Singapore around mid-2025, shutting its China offices and laying off dozens of employees. Meta assured regulators there would be \"no continuing Chinese ownership interests.\" Beijing wasn't buying it. The National Development and Reform Commission \x3Ca href=\"https://www.bloomberg.com/news/articles/2026-04-27/china-blocks-meta-s-2-billion-acquisition-of-ai-startup-manus\" rel=\"noopener\" target=\"_blank\">issued a one-line decree\x3C/a> blocking the deal and ordering both parties to withdraw. Chinese officials reportedly described the acquisition as a \x3Ca href=\"https://www.cnbc.com/2026/04/28/china-meta-manus-ai-deal.html\" rel=\"noopener\" target=\"_blank\">\"conspiratorial\" attempt\x3C/a> to hollow out the country's technology base. The decision was elevated to China's National Security Commission - the body chaired by Xi Jinping himself. The move comes alongside a \x3Ca href=\"https://www.bloomberg.com/news/articles/2026-04-29/china-s-meta-backlash-renders-manus-model-officially-dead\" rel=\"noopener\" target=\"_blank\">broader crackdown\x3C/a> barring ByteDance, Moonshot AI, and other firms from taking American capital without approval. Bloomberg declared the \"Manus model\" of Chinese startups relocating overseas to attract Western buyers \x3Ca href=\"https://www.bloomberg.com/news/articles/2026-04-29/china-s-meta-backlash-renders-manus-model-officially-dead\" rel=\"noopener\" target=\"_blank\">\"officially dead.\"\x3C/a>"},$R[439]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>The AI cold war now has its clearest red line yet. Simply moving your headquarters to Singapore doesn't get you out of Beijing's reach if your founders, technology, or research ecosystem trace back to China. For founders and VCs, this changes the calculus on every cross-border AI deal. And for Meta specifically, it's a missed opportunity to leapfrog in the \x3Ca href=\"https://techcrunch.com/2026/04/27/china-vetoes-metas-2b-manus-deal-after-months-long-probe/\" rel=\"noopener\" target=\"_blank\">agentic AI race\x3C/a> against Google and OpenAI. The global AI industry is bifurcating in real time - and neither side is willing to let talent or technology cross the line."},$R[440]={t:"h2",c:"\x3Cstrong>ROGO RAISES $160M TO REPLACE YOUR JUNIOR BANKER\x3C/strong>"},$R[441]={t:"img",c:"/archive-images/5nw9T5LUVQ4FiX2VSap9es6bvc-04739ffa.avif"},$R[442]={t:"p",c:"Image Source: \x3Ca href=\"https://rogo.ai/news/series-d\" rel=\"noopener\" target=\"_blank\">Rogo\x3C/a>"},$R[443]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>Rogo, the AI platform built specifically for investment banking, closed a $160 million Series D led by 
1Kleiner Perkins on Tuesday - bringing total funding past $300 million. Its AI agent Felix now handles deal screening, CIM generation, buyer outreach, and data room diligence autonomously."},$R[444]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>Founded by Gabriel Stengel, John Willett, and Tumas Rackaitis - three Princeton classmates who left jobs at \x3Ca href=\"https://techfundingnews.com/rogo-160m-series-d-kleiner-perkins-investment-banking-ai/\" rel=\"noopener\" target=\"_blank\">J.P. Morgan and Lazard\x3C/a> to build what started as a thesis project on econometrics chatbots. That thesis is now used by \x3Ca href=\"https://www.prnewswire.com/news-releases/rogo-raises-160m-series-d-to-scale-the-agentic-platform-for-finance-302756546.html\" rel=\"noopener\" target=\"_blank\">over 35,000 financial professionals\x3C/a> at 250+ institutions including Rothschild, Jefferies, Lazard, Moelis, and Nomura. The round included Sequoia, Thrive Capital, Khosla Ventures, J.P. Morgan Growth Equity Partners, and BoxGroup. Rogo's valuation jumped to \x3Ca href=\"https://www.tamradar.com/funding-rounds/rogo-series-d-160m\" rel=\"noopener\" target=\"_blank\">$2 billion\x3C/a> - up from $750 million just four months ago after its Series C. The key differentiator: Rogo doesn't do general-purpose AI. Its models are trained exclusively on financial workflows and integrate directly with internal data sources. Kleiner Perkins partner Mamoon Hamid called it a platform that's becoming \x3Ca href=\"https://www.prnewswire.com/news-releases/rogo-raises-160m-series-d-to-scale-the-agentic-platform-for-finance-302756546.html\" rel=\"noopener\" target=\"_blank\">\"the operating system for an entire industry.\"\x3C/a> But some industry insiders \x3Ca href=\"https://www.pymnts.com/news/investment-tracker/2026/rogo-raises-160-million-to-lessen-wall-street-workloads/\" rel=\"noopener\" target=\"_blank\">worry it could eliminate\x3C/a> the junior banker role entirely."},$R[445]={t:"p",c:"Why it matters: This is the vertical AI thesis playing out in real time. Generic copilots hit a wall in regulated industries because compliance teams can't audit a general model on Tuesday and trust it on Friday. Rogo is the second nine-figure round this week built on that insight - Aidoc raised $150M for hospital diagnosis. The generative AI market in banking is projected to grow from $1.9B to $18.5B by 2034 . The winners won't be the biggest models. They'll be the ones that own the workflow in a single industry. If you're building AI tools, the lesson is clear - go narrow, go deep, and own the data pipeline that no horizontal player can replicate."},$R[446]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.00.12_PM-ef59bfe3.png"},$R[447]={t:"h2",c:"\x3Cstrong>Quick News:\x3C/strong>"},$R[448]={t:"ul",c:$R[449]=["Meta beat Q1 earnings - then the stock tanked 8%. Revenue hit $56.31 billion , up 33% year over year. Net income nearly doubled. But Meta raised its 2026 AI capex guidance to $125-145 billion - up from $115-135B - and investors ran. Zuckerberg said Meta is \"on track to deliver personal superintelligence to billions of people.\" Meanwhile, 8,000 layoffs start May 20.","600+ Google employees demanded Pichai reject Pentagon AI work. The letter, led by DeepMind researchers , cited concerns over lethal autonomous weapons and mass surveillance. Google signed the $200 million deal anyway . The contract lets the DOD use Gemini for \"any lawful purpose\" - the same language that got Anthropic blacklisted.","\x3Cstrong>NVIDIA dropped Nemotron 3 Nano Omni.\x3C/strong> The new \x3Ca href=\"https://www.cryptointegrat.com/p/ai-news-april-29-2026\" rel=\"noopener\" target=\"_blank\">30B-parameter open multimodal model\x3C/a> ships with 256K context length and claims the highest efficiency-to-accuracy ratio in its class. NVIDIA's pitch: frontier-quality AI that runs on edge devices.","\x3Cstrong>The White House is clearing a path for federal agencies to use Anthropic's Mythos.\x3C/strong> New guidance would let agencies \x3Ca href=\"https://www.cryptointegrat.com/p/ai-news-april-29-2026\" rel=\"noopener\" target=\"_blank\">bypass Anthropic's supply chain risk designation\x3C/a> and onboard the model that can autonomously discover zero-day vulnerabilities. The NSA is already using it.","The EU AI Act Omnibus talks collapsed. The April 28 trilogue in Strasbourg ended without agreement . The August 2, 2026 high-risk deadline for AI hiring tools remains unchanged. Penalties for violations reach EUR 35M or 7% of worldwide turnover."]},$R[450]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.01.58_PM-b002921b.png"},$R[451]={t:"h2",c:"\x3Cstrong>Everything in one place:\x3C/strong>"}
1,$R[452]={t:"p",c:"🧰 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/ai-toolkit\" rel=\"noopener\" target=\"_blank\">AI Toolkit\x3C/a>\x3C/strong> - our curated stack of the best AI tools"},$R[453]={t:"p",c:"🛒 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/marketplace\" rel=\"noopener\" target=\"_blank\">Marketplace\x3C/a>\x3C/strong>\x3Cstrong> \x3C/strong>- prompts, templates, and workflows"},$R[454]={t:"p",c:"📚 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/archive\" rel=\"noopener\" target=\"_blank\">Archive\x3C/a>\x3C/strong> - every past edition in one place"},$R[455]={t:"p",c:"💸 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/referrals\" rel=\"noopener\" target=\"_blank\">Referrals\x3C/a>\x3C/strong> - refer friends, earn real rewards"},$R[456]={t:"p",c:"🎨 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/creator-portal\" rel=\"noopener\" target=\"_blank\">Creators\x3C/a>\x3C/strong>\x3Cstrong> \x3C/strong>- build and sell in our ecosystem"},$R[457]={t:"p",c:"✍️ \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/write-for-us\" rel=\"noopener\" target=\"_blank\">Write for Us\x3C/a>\x3C/strong> - share your AI expertise with 25K readers"},$R[458]={t:"p",c:"📣 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/advertise\" rel=\"noopener\" target=\"_blank\">Advertise\x3C/a>\x3C/strong> - reach 25,000+ AI professionals"},$R[459]={t:"img",c:"/archive-images/community_wins__1_-85006d73.png"},$R[460]={t:"h2",c:"\x3Cstrong>Share the AI Debrief. Get paid. \x3C/strong>"},$R[461]={t:"p",c:"\x3Cstrong>Refer 1 →\x3C/strong> your name shouted out to 25,000 readers"},$R[462]={t:"p",c:"\x3Cstrong>Refer 5 →\x3C/strong> mystery reward with real dollar value"},$R[463]={t:"p",c:"\x3Cstrong>Refer 10 →\x3C/strong> $100 back, every time. No cap. Forever."},$R[464]={t:"p",c:"Your referral link is below - one share is all it takes."},$R[465]={t:"p",c:"\x3Cstrong>Big thanks to these 102 readers who referred at least one person this week. You're the reason this newsletter keeps growing:\x3C/strong>"},$R[466]={t:"p",c:"Liam T. · Aisha R. · Marcus W. · Jenna P. · Rohan S. · Brianna K. · Caleb M. · Yuki L. · Deshawn F. · Mia G. · Oscar V. · Tanya H. · Felix J. · Alana D. · Ravi N. · Morgan E. · Dante C. · Sienna B. · Kai O. · Gabriella Z. · Trevor A. · Priyanka M. · Blake R. · Lucia S. · Jamal K. · Heather W. · Soren P. · Catalina F. · Wyatt D. · Nadia L. · Corey H. · Simone T. · Arjun B. · Kelsey J. · Ezra N. · Tatiana G. · Malik V. · Jocelyn C. · Theo R. · Marisol A. · Dominic E. · Harper K. · Idris M. · Cassidy W. · Nolan S. · Zara P. · Jared F. · Lydia H. · Roman D. · Aileen T. · Preston B. · Noelle L. · Kwame J. · Savannah R. · Emilio G. · Tessa K. · Darius N. · Paloma C. · Griffin M. · Annika S. · Cedric W. · Joanna F. · Mateo H. · Sloane D. · Tariq A. · Bethany P. · Nico L. · Imani R. · Gavin T. · Renata K. · Sterling B. · Maren J. · Alejandro V. · Piper G. · Rashid E. · Courtney M. · Landon S. · Esme W. · Jerome F. · Adriana H. · Bennett D. · Yara T. · Maxwell C. · Ingrid R. · Tobias N. · Serena L. · Colton A. · Farrah K. · Santiago P. · Delaney B. · Amir J. · Molly G. · Tristan V. · Kendra M. · Hugo S. · Camille W. · Desmond F. · Iris D. · Xavier T. · Reese H. · Joaquin R. · Elliot K."},$R[467]={t:"p",c:"\x3Cstrong>Want your name here next week? One referral is all it takes.\x3C/strong>"},$R[468]={t:"p",c:"- Drew & the rest of the humans behind The AI Debrief"}],edition:"toolkit",issueNo:null},$R[469]={slug:"the-ai-arms-race-just-got-personal",title:"The AI Arms Race Just Got Personal",subtitle:"OpenAI doubles the price of intelligence. SpaceX bets $60B on coding. And someone broke into the most dangerous AI ever built.",date:"2026-04-24",thumbnail:"/archive-images/1_v2i15z59zTun66sjhgTURQ-b7cbb3af.png",blocks:$R[470]=[$R[471]={t:"p",c:"\x3Cstrong>In partnership with\x3C/strong>"},$R[472]={t:"img",c:"/archive-images/logo-full-dark-7e6ffaa4.png"},$R[473]={t:"h3",c:"Talk to your AI tools the way you'd talk to a colleague."},$R[474]={t:"img",c:"/archive-images/flow-works-anywhere-green-f750ad70.png"},$R[475]={t:"p",c:"You don't send a colleague a three-word brief. You explain the context, the constraints, what you've already tried. But typing all that into ChatGPT takes forever — so you don't."},$R[476]={t:"p",c:"\x3Ca href=\"https://ref.wisprflow.ai/beehiiv-ai/?_bhiiv=opp_0ef3b2bf-b6f1-490d-aa36-12790dcc5a9d_4de8c0ec&amp;bhcl_id=4f3a0892-77ce-4a3b-921e-e8f996e55951_%7B%7Bsubscriber_id%7D%7D_%7B%7Bemail_address_id%7D%7D\" rel=\"noopener\" target=\"_blank\">Wispr Flow\x3C/a> lets you speak your prompts instead. Talk through your thinking naturally and get clean, paste-ready text. No filler words. No cleanup. Just detailed prompts that actually get you useful answers on the first try."},$R[477]={t:"p",c:"Millions of users worldwide. Works system-wide on Mac, Windows, and iPhone."},$R[478]={t:"p",c:"\x3Ca href=\"https://ref.wisprflow.ai/beehiiv-ai/?_bhiiv=opp_0ef3b2bf-b6f1-490d-aa36-12790dcc5a9d_4de8c0ec&amp;bhcl_id=4f3a0892-77ce-4a3b-921e-e8f996e55951_%7B%7Bsubscriber_id%7D%7D_%7B%7Bemail_address_id%7D%7D\" rel=\"noopener\" target=\"_blank\">Try Wispr Flow free\x3C/a>"},$R[479]={t:"img",c:"/archive-images/Screenshot_2026-03-26_at_10.04.45_PM-5872dae0.png"}
1,$R[480]={t:"p",c:"OpenAI just dropped GPT-5.5 and called it a \"new class of intelligence.\" Elon Musk's rocket company might buy the hottest coding startup on Earth for $60 billion. And the most dangerous AI model ever built? Someone already broke into it. Let's get into it."},$R[481]={t:"p",c:"\x3Cstrong>In today’s AI debrief:\x3C/strong>"},$R[482]={t:"ul",c:$R[483]=["OpenAI launches GPT-5.5 - its first full retrain since GPT-4.5","SpaceX locks a $60B option to acquire Cursor","Anthropic investigates unauthorized access to its Mythos model"]},$R[484]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_5.51.07_PM-fab03707.png"},$R[485]={t:"h2",c:"\x3Cstrong>GPT-5.5 IS HERE - AND IT'S NOT CHEAP\x3C/strong>"},$R[486]={t:"img",c:"/archive-images/upload-1776986463153-c0b6bf66-657a812a.avif"},$R[487]={t:"p",c:"Image Source: \x3Ca href=\"https://www.chatai.com/\" rel=\"noopener\" target=\"_blank\">chatai.com\x3C/a>"},$R[488]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>OpenAI released GPT-5.5 on Thursday - its first completely retrained base model since GPT-4.5 and the latest step in a sprint toward building an AI \"super app.\" The model is rolling out to paid ChatGPT subscribers and Codex users now, with API access coming soon."},$R[489]={t:"p",c:"The details: Every GPT-5.x model between 4.5 and 5.5 was a post-training iteration on the same base. GPT-5.5 is a ground-up rebuild - new architecture, new training data, new agent-oriented objectives. The results show it. On Terminal-Bench 2.0 , which tests complex command-line workflows, GPT-5.5 scored 82.7% compared to Claude Opus 4.7 at 69.4% and Gemini 3.1 Pro at 68.5%. It tops the Artificial Analysis Intelligence Index at 60 - three points ahead of the competition. The model ships with a 1M-token context window and matches GPT-5.4's per-token speed despite being significantly more capable. OpenAI President Greg Brockman called it \"a new class of intelligence\" and framed it as another step toward a unified super app combining ChatGPT, Codex, and an AI browser. ChatGPT now has over 900 million weekly active users and 50 million subscribers . But the pricing tells the real story: API costs double from GPT-5.4 at $5 per million input tokens and $30 per million output - the largest single-release price jump in the GPT-5.x series. OpenAI argues the model uses roughly 40% fewer tokens to complete the same tasks, netting out to about 20% higher total cost for equivalent work."},$R[490]={t:"p",c:"Why it matters: The model race just entered a new phase. Six weeks between GPT-5.4 and GPT-5.5. Less than two months since Anthropic dropped Claude Mythos Preview . The gap between releases is shrinking while the capabilities keep climbing. For developers and businesses, the question is no longer \"which model is best\" - it's whether you can keep up with the upgrade cycle. And with GPT-5.5 doubling API costs, the era of cheap frontier AI is officially over. The companies that figure out how to extract real ROI from these tools - not just test them - will pull ahead fast."},$R[491]={t:"h2",c:"\x3Cstrong>SPACEX LOCKS $60 BILLION CURSOR DEAL\x3C/strong>"},$R[492]={t:"img",c:"/archive-images/gettyimages-2271955278-390cfe2a.webp"},$R[493]={t:"p",c:"Image Source: \x3Ca href=\"http://cnet.com\" rel=\"noopener\" target=\"_blank\">cnet.com\x3C/a>"},$R[494]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>SpaceX announced on Tuesday that it has secured the right to acquire Cursor - the AI coding tool that's been eating the developer market alive - for $60 billion later this year, or pay $10 billion for the collaboration. The deal killed a $2 billion fundraise that was about to close."},$R[495]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>\x3Ca href=\"https://techcrunch.com/2026/04/22/how-spacex-preempted-a-2b-fundraise-with-a-60b-buyout-offer/\" rel=\"noopener\" target=\"_blank\">Cursor was days away\x3C/a> from closing a $2 billion funding round when SpaceX swooped in with a $10 billion \"collaboration fee\" and a path to full acquisition. The startup halted investor discussions immediately. The logic: that $2B raise still wouldn't have gotten Cursor to cash-flow breakeven, meaning another capital raise was inevitable. SpaceX offered something better - \x3Ca href=\"https://www.cnbc.com/2026/04/21/spacex-says-it-can-buy-cursor-later-this-year-for-60-billion-or-pay-10-billion-for-our-work-together.html\" rel=\"noopener\" target=\"_blank\">access to xAI's Colossus supercomputer\x3C/a> with its million-H100-equivalent compute capacity. Cursor CEO Michael Truell, a 25-year-old former Google intern, said he's \"excited to partner with the SpaceX team to scale up Composer\" - Cursor's AI model. \x3Ca href=\"https://www.cnbc.com/2026/04/22/microsoft-looked-at-buying-cursor-before-spacex-deal-sources-say.html\" rel=\"noopener\" target=\"_blank\">Microsoft had looked at a deal\x3C/a> for Cursor before deciding not to proceed, while GitHub Copilot continues losing ground. The timing is strategic: SpaceX is \x3Ca href=\"https://techcrunch.com/2026/04/22/how-spacex-preempted-a-2b-fundraise-with-a-60b-buyout-offer/\" rel=\"noopener\" target=\"_blank\">targeting a June IPO\x3C/a> at a $1.75-1.8 trillion valuation and wants investors to see it as more than a space and satellite business. The actual acquisition would come after the IPO to avoid updating confidential financial filings."},$R[496]={t:"p",c:"Why it matters: This deal tells you exactly where the AI money is going: agentic coding . Cursor was valued at $2.5 billion in January 2025. Now it's worth $60 billion - a 24x jump in 15 months. That's not hype, that's developers voting with their wallets for tools that actually ship code. But here's the tension: neither Cursor nor xAI has proprietary models that can match Anthropic or OpenAI. SpaceX has compute. Cursor has distribution. What they're betting on is that the combination creates s
1omething neither could build alone. If you're a founder or builder, the signal is clear - AI coding isn't a feature anymore. It's the product category that every major player in tech is willing to pay tens of billions to own."},$R[497]={t:"h2",c:"\x3Cstrong>ANTHROPIC'S MYTHOS JUST GOT BREACHED\x3C/strong>"},$R[498]={t:"img",c:"/archive-images/gettyimages-2271848057-3587cc46.webp"},$R[499]={t:"p",c:"Image Source: \x3Ca href=\"http://cbsnews.com\" rel=\"noopener\" target=\"_blank\">cbsnews.com\x3C/a>"},$R[500]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>Anthropic confirmed on Wednesday that it's investigating \"unauthorized access\" to Claude Mythos Preview - the AI model it deliberately kept from the public because of its ability to find and exploit zero-day vulnerabilities across every major operating system and web browser. A small group accessed the model through a third-party contractor portal."},$R[501]={t:"p",c:"The details: When Anthropic launched Mythos on April 7 under \"Project Glasswing,\" it gave access to only a handful of trusted companies - Amazon, Apple, Google, Microsoft, JPMorgan Chase, Nvidia, and Cisco - to help patch the thousands of high-severity vulnerabilities the model had discovered. The UK's AI Security Institute tested Mythos and found it could complete multi-stage corporate network attacks - a 32-step simulation spanning initial reconnaissance to full network takeover that would take human security professionals days. It succeeded 73% of the time on expert-level tasks that no previous model could complete. Now, Bloomberg reported that a group gained access through a third-party vendor environment using internet sleuthing tools. Anthropic says it hasn't detected breaches beyond the vendor environment, and the group reportedly wanted to try the models rather than use them maliciously. Security researcher Bruce Schneier wrote this week that the implications are significant: the model can autonomously discover and weaponize software vulnerabilities without expert guidance. Meanwhile, the NSA has reportedly started using Mythos , and Anthropic CEO Dario Amodei visited the White House last week to discuss broader access."},$R[502]={t:"p",c:"Why it matters: The irony is brutal. The AI model built to find security flaws had a security flaw in its own access controls. But the bigger story is what this means for every company and individual running software. Mythos can find vulnerabilities that thousands of human developers missed . If this capability proliferates - whether through leaks, competitor models, or open-source alternatives - the cybersecurity landscape changes permanently. The World Economic Forum warned that even with defensive use, organizations may face \"overload\" from thousands of newly discovered vulnerabilities they can't patch fast enough. We're entering an era where AI doesn't just assist hackers or defenders - it fundamentally changes the math of who has the advantage."},$R[503]={t:"img",c:"/archive-images/go_deeper-5e7b7ab8.png"},$R[504]={t:"h2",c:"\x3Cstrong>You’re reading about AI. AI Insider members are using it. \x3C/strong>"},$R[505]={t:"p",c:"Every week I go live and break down one AI tool or workflow - exactly how I'm using it, what's working, what's not. You ask questions, I answer live. Everything gets saved to the vault."},$R[506]={t:"p",c:"Join today and you unlock instantly:"},$R[507]={t:"ul",c:$R[508]=["20+ recorded live sessions in the vault","Weekly live breakdowns with Q&A","AI tools database + step-by-step courses","Early access to new AI companies","300+ founders and builders doing the same thing"]},$R[509]={t:"p",c:"Total value: $5000+ - yours for $97/month"},$R[510]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.00.12_PM-ef59bfe3.png"},$R[511]={t:"ul",c:$R[512]=["\x3Cstrong>Google confirmed Gemini will power the new Siri.\x3C/strong> At Cloud Next 2026 in Las Vegas, Google Cloud CEO Thomas Kurian \x3Ca href=\"https://www.macrumors.com/2026/04/22/google-gemini-powered-siri-2026/\" rel=\"noopener\" target=\"_blank\">stood in front of an Apple logo\x3C/a> and said Gemini technology will power \"a more personalized Siri coming later this year.\" The updated assistant is expected to debut with \x3Ca href=\"https://9to5mac.com/2026/04/22/google-teases-gemini-powered-siri-upgrade-
1during-cloud-next-keynote/\" rel=\"noopener\" target=\"_blank\">iOS 27 this fall\x3C/a> and include a standalone Siri app with chat logs and multi-turn conversations. Two historic rivals, now building the same product.","Meta is tracking employee keystrokes to train AI. The company installed software on U.S. employees' work computers that captures mouse movements, clicks, keystrokes, and screenshots across hundreds of websites including Google, LinkedIn, and GitHub. Multiple employees called it \"dystopian\" in internal messages. The goal: build AI agents that can do white-collar work autonomously. This comes as Meta plans to lay off 10% of its global workforce starting May 20 .","\x3Cstrong>Geoffrey Hinton told the UN to hit the brakes on AI.\x3C/strong> The Nobel laureate and \"godfather of AI\" \x3Ca href=\"https://news.un.org/en/story/2026/04/1167361\" rel=\"noopener\" target=\"_blank\">told delegates at the Digital World Conference\x3C/a> that AI is \"a very fast car with no steering wheel\" and called for immediate regulation before capabilities outpace governance entirely.","\x3Cstrong>Spotify launched voluntary AI labels in beta.\x3C/strong> Listeners can now see whether AI was involved in a song's \x3Ca href=\"https://www.digitalmusicnews.com/2026/04/22/spotify-tests-ai-credit-labels-distrokid/\" rel=\"noopener\" target=\"_blank\">vocals, lyrics, or production\x3C/a> directly in Song Credits. The catch: it's entirely voluntary, starting with DistroKid artists. Apple Music already requires its own \"Transparency Tags\" - but those rely on accurate self-reporting too.","\x3Cstrong>Google unveiled TPU 8 chips at Cloud Next.\x3C/strong> The new dual-chip approach - \x3Ca href=\"https://www.techi.com/google-tpu-8-nvidia-competition/\" rel=\"noopener\" target=\"_blank\">TPU 8t for training, TPU 8i for inference\x3C/a> - claims 2.7x better price-to-performance than the prior generation. Broadcom co-designed the training chip; MediaTek handled inference. Google is now the only hyperscaler selling custom AI chips externally, and Nvidia's monopoly narrative just got a lot harder to maintain."]},$R[513]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.01.58_PM-b002921b.png"},$R[514]={t:"h2",c:"\x3Cstrong>Everything in one place:\x3C/strong>"},$R[515]={t:"p",c:"🧰 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/ai-toolkit\" rel=\"noopener\" target=\"_blank\">AI Toolkit\x3C/a>\x3C/strong> - our curated stack of the best AI tools"},$R[516]={t:"p",c:"🛒 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/marketplace\" rel=\"noopener\" target=\"_blank\">Marketplace\x3C/a>\x3C/strong>\x3Cstrong> \x3C/strong>- prompts, templates, and workflows"},$R[517]={t:"p",c:"📚 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/archive\" rel=\"noopener\" target=\"_blank\">Archive\x3C/a>\x3C/strong> - every past edition in one place"},$R[518]={t:"p",c:"💸 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/referrals\" rel=\"noopener\" target=\"_blank\">Referrals\x3C/a>\x3C/strong> - refer friends, earn real rewards"},$R[519]={t:"p",c:"🎨 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/creator-portal\" rel=\"noopener\" target=\"_blank\">Creators\x3C/a>\x3C/strong>\x3Cstrong> \x3C/strong>- build and sell in our ecosystem"},$R[520]={t:"p",c:"✍️ \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/write-for-us\" rel=\"noopener\" target=\"_blank\">Write for Us\x3C/a>\x3C/strong> - share your AI expertise with 25K readers"},$R[521]={t:"p",c:"📣 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/advertise\" rel=\"noopener\" target=\"_blank\">Advertise\x3C/a>\x3C/strong> - reach 25,000+ AI professionals"},$R[522]={t:"img",c:"/archive-images/community_wins-70bb6569.png"},$R[523]={t:"h2",c:"\x3Cstrong>This week’s top referrer:\x3C/strong>"},$R[524]={t:"p",c:"\x3Cstrong>Marcus T.\x3C/strong> - Product Manager, Chicago IL"},$R[525]={t:"p",c:"Marcus found The AI Debrief through a coworker's Slack message three weeks ago. Since then he's referred 14 readers and earned $100 back - twice."},$R[526]={t:"p",c:"\"I dropped the link in our team channel and it just spread. Now half my product org reads it before standup on Frid
1ays.\""},$R[527]={t:"p",c:"These are the readers building with AI - not just reading about it."},$R[528]={t:"p",c:"\x3Cstrong>Big thanks to these 50 readers who referred at least one person this week. You're the reason this newsletter keeps growing:\x3C/strong>"},$R[529]={t:"p",c:"Priya S. · Jordan K. · David L. · Amanda R. · Carlos M. · Sofia C. · Ryan H. · Nina P. · Derek W. · Emily F. · Rachel G. · Tom B. · Jasmine A. · Kevin N. · Laura D. · Brandon T. · Vanessa Q. · Eric Y. · Hannah J. · Mike Z. · Olivia R. · Tyler S. · Maya I. · Chris E. · Alyssa V. · Nathan B. · Grace L. · Samira K. · Patrick O. · Elena T. · Zach M. · Danielle H. · Andre C. · Megan F. · Isaiah W. · Leah P. · Connor D. · Fatima N. · Justin R. · Brooke A. · Raj M. · Stephanie L. · Aaron K. · Chloe T. · Diego S. · Natasha B. · Owen J. · Kira H. · Sean O. · Amara V."},$R[530]={t:"p",c:"\x3Cstrong>Want your name here next week? One referral is all it takes.\x3C/strong>"},$R[531]={t:"img",c:"/archive-images/community_wins__1_-85006d73.png"},$R[532]={t:"h2",c:"\x3Cstrong>Share the AI Debrief. Get paid. \x3C/strong>"},$R[533]={t:"p",c:"\x3Cstrong>Refer 1 →\x3C/strong> your name shouted out to 25,000 readers"},$R[534]={t:"p",c:"\x3Cstrong>Refer 5 →\x3C/strong> mystery reward with real dollar value"},$R[535]={t:"p",c:"\x3Cstrong>Refer 10 →\x3C/strong> $100 back, every time. No cap. Forever."},$R[536]={t:"p",c:"Your referral link is below - one share is all it takes."},$R[537]={t:"p",c:"P.S. - Jonathan from Wisconsin just hit 10 referrals and got $100 back."},$R[538]={t:"p",c:"You're one share away from seeing your name in next week's edition - 25,000 people will see it."},$R[539]={t:"p",c:"- Drew & the rest of the humans behind The AI Debrief"}],edition:"toolkit",issueNo:null},$R[540]={slug:"florida-just-came-for-openai",title:"Florida Just Came For OpenAI",subtitle:"A state AG, a $122B round, and the outage that rattled every enterprise buyer.",date:"2026-04-16",thumbnail:"/archive-images/hed_image__1_-92d6e2d8.png",blocks:$R[541]=[$R[542]={t:"img",c:"/archive-images/Screenshot_2026-03-26_at_10.04.45_PM-5872dae0.png"},$R[543]={t:"p",c:"This week, AI stopped being a story about models and started being a story about consequences. A state attorney general filed subpoenas. The most valuable private company in history closed the biggest round in venture history. And the AI your team depends on went dark - on purpose or not, everyone noticed."},$R[544]={t:"p",c:"\x3Cstrong>In today’s AI debrief:\x3C/strong>"},$R[545]={t:"ul",c:$R[546]=["Florida's AG just opened the first state-level criminal-adjacent probe into OpenAI - and it could reshape AI liability forever","Anthropic is raising at an $800B valuation, more than doubling its February mark","A Claude outage exposed the hidden fragility in every \"AI-first\" company's stack"]},$R[547]={t:"p",c:"Let's get into it."},$R[548]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_5.51.07_PM-fab03707.png"},$R[549]={t:"h2",c:"\x3Cstrong>Florida AG Opens Criminal-Adjacent Probe Into OpenAI\x3C/strong>"},$R[550]={t:"img",c:"/archive-images/1775757337460-6a94cde3.webp"},$R[551]={t:"p",c:"Image source: \x3Ca href=\"https://www.axios.com/2026/04/09/florida-ag-launches-investigation-openai\" rel=\"noopener\" target=\"_blank\">Axios\x3C/a>"},$R[552]={t:"p",c:"The debrief: Florida Attorney General James Uthmeier launched a formal investigation into OpenAI last Thursday, citing the FSU mass shooting, child safety, and national security - and subpoenas are already on the way ."},$R[553]={t:"p",c:"The details: Lawyers for one of the FSU victims claim the shooter was in constant communication with ChatGPT before the attack , including guidance on firearm use. The probe covers three distinct threads: chatbot involvement in the shooting, generation of child sexual abuse material, and whether OpenAI's data could be exploited by foreign adversaries. OpenAI says it will cooperate, and has already published a Child Safety Blueprint in response to mounting pressure. The timing is brutal - OpenAI is reportedly preparing for an IPO at a potential $1T valuation ."},$R[554]={t:"p",c:"Why it matters: This is the template. Every founder shipping AI to consumers should read Uthmeier's statement as a preview of what's coming in the other 49 states. The Trump administration has taken a hands-off federal approach , which means enforcement will now come state-by-state - and the legal question at the center of it (\"can an AI tool be held liable for what users do with it?\") will define product design for the next decade. If you're building anything consumer-facing, your safety layer just became a fundraising line item."},$R[555]={t:"h2",c:"\x3Cstrong>Anthropic Eyes $800B - More Than Doubling Its Feb Valuation\x3C/strong>"},$R[556]={t:"img",c:"/archive-images/cf9d0c0b-497f-4847-ba80-56ec27d6c7b4-88e14949.jpeg"},$R[557]={t:"p",c:"Image source: \x3Ca href=\"https://uk.finance.yahoo.com/news/anthropic-attracts-investor-offers-at-an-800-billion-valuation-000636800.html\" rel=\"noopener\" target=\"_blank\">Yahoo\x3C/a>"},$R[558]={t:"p",c:"The debrief: Anthropic is in talks to raise at an $800 billion valuation , up from $380B just two months ago - while OpenAI officially closed its $122B round at an $852B valuation , the largest private round in history."},$R[559]={t:"p",c:"The details: Anthropic's annualized revenue jumped from $9B to $19B in just three months , fastest relative growth of any frontier lab. OpenAI's round was led by Amazon ($50B, though $35B is contingent on IPO or AGI), SoftBank ($30B), and Nvidia ($30B). Four deals - OpenAI, Anthropic, xAI, and Waymo - accounted for 65% of all global venture investment in Q1 2026 . Meanwhile, Upscale AI is closing a $200M round at a $2B valuation to build infrastru
1cture connecting AI compute clusters - a reminder that the shovels in this gold rush are still commanding premium prices."},$R[560]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>The \x3Ca href=\"https://www.pwc.com/gx/en/news-room/press-releases/2026/pwc-2026-ai-performance-study.html\" rel=\"noopener\" target=\"_blank\">AI economy is consolidating around fewer and fewer companies\x3C/a> - PwC's new study shows 74% of AI's economic value is already captured by just 20% of organizations. But here's the uncomfortable part: Anthropic's valuation more than doubled in 60 days without a major model release. That's not product-led growth - that's scarcity-led growth. The capital is chasing the last remaining seats at the frontier table, and if you're a founder raising an AI-adjacent round right now, you have a window that won't last through Q3."},$R[561]={t:"h2",c:"\x3Cstrong>The Claude Outage That Exposed Every \"AI-First\" Stack\x3C/strong>"},$R[562]={t:"img",c:"/archive-images/claude_down_error-4f71a7a0.webp"},$R[563]={t:"p",c:"Image Source: \x3Ca href=\"https://www.bleepingcomputer.com/news/artificial-intelligence/anthropic-confirms-claude-is-down-in-a-worldwide-outage/\" rel=\"noopener\" target=\"_blank\">BleepingComputer\x3C/a>"},$R[564]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>Claude, Claude Code, and the Anthropic API all \x3Ca href=\"https://blog.greeden.me/en/2026/04/16/week-2-of-april-2026-one-week-generative-ai-news-roundup-what-gpt-5-4-cyber-claude-cyber-concerns-and-gemini-notebooks-show-about-what-comes-next-for-practical-ai/\" rel=\"noopener\" target=\"_blank\">went down this week\x3C/a> - and thousands of businesses running production workflows on top of it went down with them."},$R[565]={t:"p",c:"The details: The outage hit during peak US working hours and affected everything from customer support agents to code-shipping pipelines at startups that had bet their entire workflow on a single model provider. This landed in the same week OpenAI announced GPT-5.4-Cyber and expanded its Trusted Access for Cyber program , signaling that reliability and identity verification are becoming the real product differentiators - not parameter counts. Google's Gemini Notebooks rollout is pushing the same message: structured, project-based AI beats ad-hoc chat."},$R[566]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>If your company depends on AI to function, single-provider reliance is now an unacceptable risk. Smart teams are already building model-agnostic routing layers so they can failover from Claude to GPT to Gemini in real time. The differentiator in 2026 isn't which model you use - it's whether your stack survives when one goes down. Enterprise buyers asked two questions this week that they didn't ask last month: \"What's your failover plan?\" and \"What's your SLA?\" Have answers ready."},$R[567]={t:"img",c:"/archive-images/go_deeper-5e7b7ab8.png"},$R[568]={t:"h2",c:"\x3Cstrong>You’re reading about AI. AI Insider members are using it. \x3C/strong>"},$R[569]={t:"p",c:"Every week I go live and break down one AI tool or workflow - exactly how I'm using it, what's working, what's not. You ask questions, I answer live. Everything gets saved to the vault."},$R[570]={t:"p",c:"Join today and you unlock instantly:"},$R[571]={t:"ul",c:$R[572]=["20+ recorded live sessions in the vault","Weekly live breakdowns with Q&A","AI tools database + step-by-step courses","Early access to new AI companies","300+ founders and builders doing the same thing"]},$R[573]={t:"p",c:"Total value: $5000+ - yours for $97/month"},$R[574]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.00.12_PM-ef59bfe3.png"},$R[575]={t:"ul",c:$R[576]=["Stanford's 2026 AI Index is out - US and China are razor-thin apart on model performance , and states passed a record 150 AI bills in 2025","\x3Cstrong>OpenAI is projecting $100B in annual ad revenue by 2030\x3C/strong> - \x3Ca href=\"https://www.marketingprofs.com/opinions/2026/54530/ai-update-april-10-2026-ai-news-and-views-from-the-past-week\" rel=\"noopener\" target=\"_blank\">chatbot ads are the next platform war\x3C/a>","4chan users accidentally discovered chain-of-thought reasoning in 2022 - a full year before Google researchers published it ( full story )","\x3Cstrong>AI caregivers are catching misdiagnoses\x3C/strong> - a son used \x3Ca href=\"https://www.humai.blog/ai-news-trends-april-2026-complete-monthly-digest/\" rel=\"noopener\" target=\"_blank\">Claude + NotebookLM to spot CT scan errors\x3C/a>
1 in his mother's cancer care, reportedly extending her life","\x3Cstrong>Shield AI raised $1.5B at $12.7B\x3C/strong> - defense AI just had \x3Ca href=\"https://www.crescendo.ai/news/latest-vc-investment-deals-in-ai-startups\" rel=\"noopener\" target=\"_blank\">its biggest round since Anduril\x3C/a>"]},$R[577]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.01.58_PM-b002921b.png"},$R[578]={t:"h2",c:"\x3Cstrong>Everything in one place:\x3C/strong>"},$R[579]={t:"p",c:"🧰 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/ai-toolkit\" rel=\"noopener\" target=\"_blank\">AI Toolkit\x3C/a>\x3C/strong> - our curated stack of the best AI tools"},$R[580]={t:"p",c:"🛒 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/marketplace\" rel=\"noopener\" target=\"_blank\">Marketplace\x3C/a>\x3C/strong>\x3Cstrong> \x3C/strong>- prompts, templates, and workflows"},$R[581]={t:"p",c:"📚 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/archive\" rel=\"noopener\" target=\"_blank\">Archive\x3C/a>\x3C/strong> - every past edition in one place"},$R[582]={t:"p",c:"💸 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/referrals\" rel=\"noopener\" target=\"_blank\">Referrals\x3C/a>\x3C/strong> - refer friends, earn real rewards"},$R[583]={t:"p",c:"🎨 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/creator-portal\" rel=\"noopener\" target=\"_blank\">Creators\x3C/a>\x3C/strong>\x3Cstrong> \x3C/strong>- build and sell in our ecosystem"},$R[584]={t:"p",c:"✍️ \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/write-for-us\" rel=\"noopener\" target=\"_blank\">Write for Us\x3C/a>\x3C/strong> - share your AI expertise with 25K readers"},$R[585]={t:"p",c:"📣 \x3Cstrong>\x3Ca href=\"https://thedebrief.ai/advertise\" rel=\"noopener\" target=\"_blank\">Advertise\x3C/a>\x3C/strong> - reach 25,000+ AI professionals"},$R[586]={t:"img",c:"/archive-images/community_wins-70bb6569.png"},$R[587]={t:"h2",c:"\x3Cstrong>This week’s top referrer:\x3C/strong>"},$R[588]={t:"p",c:"\x3Cstrong>Kelly M. - AI Enthusiast, Austin TX\x3C/strong>"},$R[589]={t:"p",c:"Kelly stumbled onto The AI Debrief a few weeks ago. She's not a founder, not an engineer - just someone who wanted to actually understand what was happening in AI without wading through 40 tabs every morning. Since subscribing, she's referred 12 readers and earned $100 back."},$R[590]={t:"p",c:"These are the readers building with AI - not just reading about it."},$R[591]={t:"p",c:"Big thanks to these 35 readers who referred at least one person this week. You're the reason this newsletter keeps growing:"},$R[592]={t:"p",c:"Marcus T. · Priya S. · David L. · Amanda R. · James O. · Sofia C. · Ryan K. · Nina P. · Derek H. · Emily W. · Carlos M. · Rachel G. · Tom B. · Jasmine A. · Kevin N. · Laura D. · Brandon F. · Vanessa Q. · Eric Y. · Hannah J. · Mike Z. · Olivia R. · Tyler S. · Maya I. · Chris E. · Alyssa V. · Nathan B. · Grace L. · Jordan P. · Samira K. · Patrick O. · Elena T. · Zach M. · Danielle H. · Andre C."},$R[593]={t:"p",c:"\x3Cstrong>Want your name here next week?\x3C/strong> One referral is all it takes."},$R[594]={t:"img",c:"/archive-images/community_wins__1_-85006d73.png"},$R[595]={t:"h2",c:"\x3Cstrong>Share the AI Debrief. Get paid. \x3C/strong>"},$R[596]={t:"p",c:"\x3Cstrong>Refer 1 →\x3C/strong> your name shouted out to 25,000 readers"},$R[597]={t:"p",c:"\x3Cstrong>Refer 5 →\x3C/strong> mystery reward with real dollar value"},$R[598]={t:"p",c:"\x3Cstrong>Refer 10 →\x3C/strong> $100 back, every time. No cap. Forever."}
1,$R[599]={t:"p",c:"Your referral link is below - one share is all it takes."},$R[600]={t:"p",c:"P.S. - Kelly just hit 10 referrals and got $100 back."},$R[601]={t:"p",c:"You're one share away from seeing your name in next week's edition - 25,000 people will see it."},$R[602]={t:"p",c:"- Drew & the rest of the humans behind The AI Debrief"}],edition:"toolkit",issueNo:null},$R[603]={slug:"anthropic-just-taught-ai-to-hack-the-world",title:"Anthropic Just Taught AI to Hack the World",subtitle:"A new Claude model found a 27-year-old vulnerability. The Pentagon still won't use it.",date:"2026-04-10",thumbnail:"/archive-images/04072026_New_Anthropic_Industry_Coalition_Racing_to_Use_Priv-c96538fb.webp",blocks:$R[604]=[$R[605]={t:"img",c:"/archive-images/Screenshot_2026-03-26_at_10.04.45_PM-5872dae0.png"},$R[606]={t:"p",c:"Anthropic had the most consequential week in its history. A new Claude model is autonomously hacking the world's most secure software to protect it. The Pentagon banned Claude from the U.S. military - and the courts just let it stand. And in the middle of all of it, Anthropic shipped more product than most companies do in a quarter."},$R[607]={t:"p",c:"Here's everything you need to know."},$R[608]={t:"p",c:"\x3Cstrong>In today’s AI debrief:\x3C/strong>"},$R[609]={t:"ul",c:$R[610]=["🔐 \x3Ca href=\"https://red.anthropic.com/2026/mythos-preview/\" rel=\"noopener\" target=\"_blank\">Claude Mythos just found a 27-year-old vulnerability nobody knew existed\x3C/a>","⚖️ \x3Ca href=\"https://news.bitcoin.com/federal-judges-deny-anthropic-relief-in-claude-military-ai-ban-set-may-oral-arguments/\" rel=\"noopener\" target=\"_blank\">The Pentagon banned Claude - and the D.C. Circuit let it stand\x3C/a>","🚀 \x3Ca href=\"https://platform.claude.com/docs/en/release-notes/overview\" rel=\"noopener\" target=\"_blank\">Sonnet 4.6, Managed Agents, and Claude Code all shipped in the same week\x3C/a>"]},$R[611]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_5.51.07_PM-fab03707.png"},$R[612]={t:"h2",c:"\x3Cstrong>Claude Mythos Just Found Bugs Nobody Knew Existed\x3C/strong>"},$R[613]={t:"img",c:"/archive-images/1_2walxdUQ5yB-2f45g9EZPA-6be3ad28.png"},$R[614]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>Anthropic quietly launched \x3Ca href=\"https://red.anthropic.com/2026/mythos-preview/\" rel=\"noopener\" target=\"_blank\">Claude Mythos Preview\x3C/a> - a new model built specifically for cybersecurity. It's not available to the public. It's invitation-only. And it just found a 27-year-old vulnerability in OpenBSD and multiple zero-days in the Linux kernel - the software running most of the world's servers."},$R[615]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>Anthropic simultaneously announced \x3Ca href=\"https://broadbandbreakfast.com/anthropic-launches-project-glasswing-to-defend-against-ai-cyberthreats/\" rel=\"noopener\" target=\"_blank\">Project Glasswing\x3C/a> - a partnership with Apple, Nvidia, Amazon, Google, and JPMorgan to use Mythos to find and fix critical vulnerabilities in their foundational systems before bad actors do. The model runs autonomously: it reads code, hypothesizes vulnerabilities, tests them, and files detailed bug reports - no human in the loop."},$R[616]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>This is the first time an AI has demonstrably matched - and likely exceeded - the world's best human security researchers at finding flaws. Every company running software (every company) is now operating in a world where AI can find the holes in their walls faster than they can patch them. The only question is whether the AI doing the finding is working for you or against you."},$R[617]={t:"h2",c:"\x3Cstrong>The Pentagon Banned Claude. The Courts Just Let It Stand.\x3C/strong>"},$R[618]={t:"img",c:"/archive-images/260326-aario-amodei-claude-anthropic-es-a0ececf0.webp"},$R[619]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>The D.C. Circuit Court of Appeals \x3Ca href=\"https://news.bitcoin.com/federal-judges-deny-anthropic-relief-in-claude-military-ai-ban-set-may-oral-arguments/\" rel=\"noopener\" target=\"_blank\">denied Anthropic's emergency request\x3C/a> to halt the Pentagon's blacklist of Claude from all U.S. military contracts. The ban stays in force while litigation continues."},$R[620]={t:"p",c:"The details: The dispute started when the Pentagon demanded Anthropic remove two things from its terms of service - a ban on fully autonomous weapons (armed drone swarms with no human oversight) and a prohibition on mass surveillance of U.S. citizens. Anthropic refused. Defense Secretary Pete Hegseth called those guardrails \"irrational obstacles.\" Anthropic called them safety requirements. The result: Claude is now designated a supply chain risk, meaning major DoD contractors, including Amazon, Microsoft, and Palantir, cannot use it. Oral arguments are set for May 19 ."},$R[621]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>This is the defining tension in AI right now - safety guardrails vs. government power. Anthropic built the most capable security model on the planet this week and was simultaneously banned by the world's largest military for refusing to remove its ethics policies. How this case resolves in May will set the precedent for every AI company's relationship with the U.S. government."},$R[622]={t:"h2",c:"\x3Cstrong>Anthropic Shipped More This Week Than Most Companies Do All Year\x3C/strong>"},$R[623]={t:"img",c:"/archive-images/1773217058700-0644d90e.jpeg"},$R[624]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>Buried under the Mythos news, Anthropic had a massive product week. \x3Ca href=\"https://platform.claude.com/docs/en/release-notes/overview\" rel=\"noopener\" target=\"_blank\">Claude Sonnet 4.6 launched\x3C/a> as their new everyday model - faster, smarter on agentic tasks, and supports a 1M token context window. \x3Ca href=\"https://platform.claude.com/docs/en/release-notes/overview\" rel=\"noopener\" target=\"_blank\">Claude Managed Agents\x3C/a> hit public beta, letting developers run Claude as a fully autonomous agent with sandboxing and built-in tools. And \x3Ca href=\"https://releasebot.io/updates/anthropic/claude-code\" rel=\"noopener\" target=\"_blank\">
1Claude Code\x3C/a> shipped major updates, including a Focus view, Google Vertex AI setup wizard, and subprocess sandboxing."},$R[625]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>The practical implication: Claude can now run longer, think deeper, and operate more autonomously than ever - all in the same week. Meanwhile, back-to-back \x3Ca href=\"https://www.techradar.com/news/live/claude-anthropic-down-outage-april-6-2026\" rel=\"noopener\" target=\"_blank\">outages on April 7 and 8\x3C/a> hit thousands of users, exposing just how dependent developers and professionals have become on a single AI platform. Anthropic fixed both within hours, but the pattern raises a real infrastructure question as demand accelerates."},$R[626]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>Claude is quietly becoming the operating system for knowledge work. Sonnet 4.6 plus Managed Agents plus Claude Code means entire workflows - research, coding, analysis, communication - can now run on Claude end to end with minimal human input. The outages were a reminder that when that's your stack, uptime isn't a feature. It's everything."},$R[627]={t:"img",c:"/archive-images/go_deeper-5e7b7ab8.png"},$R[628]={t:"h2",c:"\x3Cstrong>You’re reading about AI. \x3C/strong>"},$R[629]={t:"h2",c:"\x3Cstrong>AI Insider members are using it. \x3C/strong>"},$R[630]={t:"p",c:"Every week I go live and break down one AI tool or workflow - exactly how I'm using it, what's working, what's not. You ask questions, I answer live. Everything gets saved to the vault."},$R[631]={t:"p",c:"Join today, and you unlock instantly:"},$R[632]={t:"ul",c:$R[633]=["20+ recorded live sessions in the vault","Weekly live breakdowns with Q&A","AI tools database + step-by-step courses","Early access to new AI companies","300+ founders and builders doing the same thing"]},$R[634]={t:"p",c:"Total value: $7,482 - yours for $97/month"},$R[635]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.00.12_PM-ef59bfe3.png"},$R[636]={t:"h2",c:"\x3Cstrong>Also This Week.\x3C/strong>"},$R[637]={t:"ul",c:$R[638]=["\x3Cstrong>\x3Ca href=\"https://platform.claude.com/docs/en/release-notes/overview\" rel=\"noopener\" target=\"_blank\">Claude Haiku 3 is being retired April 19\x3C/a>\x3C/strong> - if you're building on the API, migrate to Haiku 4.5 now or your calls will break","\x3Cstrong>\x3Ca href=\"https://platform.claude.com/docs/en/release-notes/overview\" rel=\"noopener\" target=\"_blank\">Anthropic's Messages API is now on Amazon Bedrock\x3C/a>\x3C/strong> as a research preview - same request shape as the native API, running on AWS infrastructure","\x3Cstrong>\x3Ca href=\"https://broadbandbreakfast.com/anthropic-launches-project-glasswing-to-defend-against-ai-cyberthreats/\" rel=\"noopener\" target=\"_blank\">Project Glasswing partners\x3C/a>\x3C/strong> include Apple, Nvidia, Google, JPMorgan, and Palo Alto Networks - the coalition defending global software infrastructure just got an AI upgrade","\x3Cstrong>\x3Ca href=\"https://news.bitcoin.com/federal-judges-deny-anthropic-relief-in-claude-military-ai-ban-set-may-oral-arguments/\" rel=\"noopener\" target=\"_blank\">May 19\x3C/a>\x3C/strong> is the date to watch - oral arguments in the Pentagon vs. Anthropic case that will define AI's relationship with U.S. government contracts"]},$R[639]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.01.58_PM-b002921b.png"},$R[640]={t:"h2",c:"\x3Cstrong>Everything in one place:\x3C/strong>"},$R[641]={t:"p",c:"🧰 \x3Ca href=\"https://thedebrief.ai/ai-toolkit\" rel=\"noopener\" target=\"_blank\">AI Toolkit\x3C/a> - our curated stack of the best AI tools"},$R[642]={t:"p",c:"🛒 \x3Ca href=\"https://thedebrief.ai/marketplace\" rel=\"noopener\" target=\"_blank\">Marketplace\x3C/a> - prompts, templates, and workflows"},$R[643]={t:"p",c:"💸 \x3Ca href=\"https://thedebrief.ai/referrals\" rel=\"noopener\" target=\"_blank\">Referrals\x3C/a> - refer friends, earn real rewards"},$R[644]={t:"p",c:"🎨 \x3Ca href=\"https://thedebrief.ai/creator-portal\" rel=\"noopener\" target=\"_blank\">Creators\x3C/a> - build and sell in our ecosystem"},$R[645]={t:"img",c:"/archive-images/community_wins-70bb6569.png"},$R[646]={t:"h2",c:"\x3Cstrong>This week’s top referrer:\x3C/strong>"},$R[647]={t:"p",c:"\x3Cstrong>Jonathan S. - Founder, Green Bay WI\x3C/strong>"},$R[648]={t:"p",c:"Jonathan found The AI Debrief through a Google search two weeks ago. Since then he's referred 10 readers and earned $100 back."},$R[649]={t:"p",c:"\"I wasn't even looking for a newsletter. Now I read it every week and I've already sent it to everyone on my team.\""},$R[650]={t:"p",c:"These are the readers building with AI - not just reading about it."},$R[651]={t:"p",c:"First person to refer someone next week gets shouted out to 25,000 readers just like this."},$R[652]={t:"img",c:"/archive-images/community_wins__1_-85006d73.png"},$R[653]={t:"h2",c:"\x3Cstrong>Share the AI Debrief. Get paid. \x3C/strong>"},$R[654]={t:"p",c:"\x3Cstrong>Refer 1 →\x3C/strong> your name shouted out to 25,000 readers"},$R[655]={t:"p",c:"\x3Cstrong>Refer 5 →\x3C/strong> mystery reward with real dollar value"},$R[656]={t:"p",c:"\x3Cstrong>Refer 10 →\x3C/strong> $100 back, every time. No cap. Forever."},$R[657]={t:"p",c:"Your referral link is below - one share is all it takes."},$R[658]={t:"p",c:"P.S. - Jonathan from Wisconsin just hit 10 referrals and got $100 back."},$R[659]={t:"p",c:"You're one share away from seeing your name in next week's edition - 25,000 people will see it."},$R[660]={t:"p",c:"- Drew & the rest of the humans behind The AI Debrief"}],edition:"toolkit",issueNo:null},$R[661]={slug:"the-week-ai-s-security-perimeter-collapsed",title:"The Week AI's Security Perimeter Collapsed",subtitle:"Anthropic leaked its most powerful model. Mercor got hacked. OpenAI raised $122 billion. All in seven days.",date:"2026-04-01",thumbnail:"/archive-images/claude-mythos-vs-opus-coding-benchmark-e0df67f2.webp",blocks:$R[662]=[$R[663]={t:"img",c:"/archive-images/Screenshot_2026-03-26_at_10.04.45_PM-f47b1882.png"},$R[664]={t:"p",c:"In the last seven days, Anthropic accidentally exposed its most powerful unreleased model, then accidentally leaked 500,000 lines of source code. Mercor - the company that trains AI models for OpenAI and Anthropic - publicly confirmed it was breached by one of the most notorious hacking groups in the world. And OpenAI closed the largest private funding round in history at an $852 billion valuation."},$R[665]={t:"p",c:"This is the week the AI industry's security perimeter broke down - at the exact moment the stakes have never been higher."},$R[666]={t:"p",c:"\x3Cstrong>In today’s AI debrief:\x3C/strong>"},$R[667]={t:"ul",c:$R[668]=["Anthropic's Claude Mythos leak: a new model tier above Opus, unprecedented cybersecurity capabilities, a market crash, and a second breach that exposed Claude Code's entire source code","Mercor confirms 4TB breach via LiteLLM supply chain attack - Lapsus$ claims your passport, source code, and VPN credentials are now for sale on the dark web","OpenAI closes $122 billion at $852 billion valuation - the largest private funding round in history, anchored by Amazon, Nvidia, and SoftBank","Quick hits on the Anthropic injunction ruling, VC hitting a record $297B in Q1, Perplexity's data lawsuit, and the week's model wars"]},$R[669]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_5.51.07_PM-fab03707.png"},$R[670]={t:"h2",c:"\x3Cstrong>ANTHROPIC ACCIDENTALLY REVEALED ITS MOST POWERFUL MODEL EVER - AND IT POSES \"UNPRECEDENTED CYBERSECURITY RISKS\"\x3C/strong>"},$R[671]={t:"img",c:"/archive-images/GettyImages-2261864825_80d206-587c7622.webp"},$R[672]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>On March 26, a misconfiguration in Anthropic's content management system left nearly 3,000 internal files publicly accessible - including a draft blog post describing a new AI model called Claude Mythos that Anthropic describes as \x3Ca href=\"https://fortune.com/2026/03/31/anthropic-source-code-claude-code-data-leak-second-security-lapse-days-after-accidentally-revealing-mythos/\" rel=\"noopener\" target=\"_blank\">\"by far the most powerful AI model we've ever developed.\"\x3C/a> The company confirmed the leak is real, calling Mythos a \"step change\" that is currently in testing with early access customers. Cybersecurity stocks immediately flash-crashed, with CrowdStrike, Palo Alto Networks, and Zscaler each dropping 6-7%. The
1n on March 31, Anthropic accidentally leaked 500,000 lines of Claude Code's source code in a separate incident - its second major data breach in five days."},$R[673]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>The leaked draft blog post revealed several things Anthropic wasn't ready to announce. First, a new model tier called \x3Ca href=\"https://www.cnbc.com/2026/03/27/anthropic-cybersecurity-stocks-ai-mythos.html\" rel=\"noopener\" target=\"_blank\">\"Capybara\"\x3C/a> - positioned above Opus, which was previously Anthropic's most capable tier. Mythos and Capybara appear to refer to the same underlying model, with Capybara being the tier name and Mythos the specific model name. The leaked document says Capybara delivers \"dramatically higher scores\" than Claude Opus 4.6 on software coding, academic reasoning, and cybersecurity tasks. The most alarming detail is the cybersecurity capability. The leaked draft describes Mythos as \"currently far ahead of any other AI model in cyber capabilities\" - capable of autonomously discovering and exploiting software vulnerabilities at speeds that outpace human defenders. Anthropic's own draft language warned the model \"presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders.\" Stifel analyst Adam Borg put it plainly: Mythos has \"the potential to become the ultimate hacking tool, and one that can elevate any ordinary hacker into a nation-state adversary.\" As a result, Anthropic is pursuing a deliberately slow rollout - seeding the model to enterprise security teams first so defenders can use it before it reaches general availability. The draft also acknowledged Mythos is \"very expensive for us to serve, and will be very expensive for our customers to use,\" with efficiency work needed before any broad release. The second breach - 500,000 lines of Claude Code source code exposed in a release packaging error - is potentially more strategically damaging. While it didn't expose model weights or customer data, it gives competitors a blueprint for reverse-engineering Claude Code's agentic harness, and security researchers have already confirmed the leaked code contains additional evidence that a \"fast\" and \"slow\" version of Capybara is in active preparation."},$R[674]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>The Mythos story is three stories running simultaneously. The first is a capability story: Anthropic has apparently built something meaningfully more powerful than anything currently on the market, and the cybersecurity implications are severe enough that the company is privately briefing government officials. The second is a market story: the cybersecurity flash crash signals that Wall Street is now pricing AI-driven disruption of the $200B+ security industry in real time. Raymond James analyst Adam Tindle's note is worth reading - he warns that \x3Ca href=\"https://www.csoonline.com/article/4151801/leak-reveals-anthropics-mythos-a-powerful-ai-model-aimed-at-cybersecurity-use-cases.html\" rel=\"noopener\" target=\"_blank\">\"defensive approaches based on known signatures or prior threat intelligence could be pressured as AI enables continuous discovery of unknown exploits.\"\x3C/a> Traditional cybersecurity moats are starting to look fragile. The third is a trust story: Anthropic is simultaneously fighting the Pentagon over safety guardrails, privately warning governments about Mythos's dangers, and accidentally leaking both its most sensitive upcoming product and its flagship coding tool's source code in the same week. For a company whose entire brand is built on being the \"safety-first\" AI lab, these two data breaches are a serious reputational problem that its critics will use for months."},$R[675]={t:"h2",c:"\x3Cstrong>LAPSUS$ CLAIMS 4TB OF MERCOR DATA - THE AI INDUSTRY'S SUPPLY CHAIN SECURITY CRISIS JUST GOT VERY REAL\x3C/strong>"},$R[676]={t:"img",c:"/archive-images/1775063256856-cc57317d.png"},$R[677]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>Mercor - the 10 billion AI talent platform that trains models for OpenAI and Anthropic -publicly confirmed on March 31 that it was hit b
1y a supply chain cyberattack through the compromised open-source LiteLLM library. \x3Ca href=\"https://cybernews.com/security/mercor-data-breach-litelllm-supply-chain-attack/\" rel=\"noopener\" target=\"_blank\">Extortion group Lapsus\x3C/a> 10 claimed responsibility, alleging it stole 4TB of data, including 939GB of source code, a 211GB database, 3TB of video and identity verification files, and full access via the company's Tailscale VPN. Mercor became the first company to publicly acknowledge being a victim of the broader TeamPCP supply chain campaign - a distinction that set off alarm bells across the entire AI startup ecosystem."},$R[678]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>The attack didn't start with Mercor. It started with LiteLLM - an open-source API gateway with 97 million monthly downloads that lets developers route calls to over 100 different AI models, including OpenAI and Anthropic. Hacking group TeamPCP first compromised Trivy, a widely used vulnerability scanner, through a misconfigured GitHub Actions workflow. That gave them the PyPI publishing credentials for LiteLLM, and they pushed two malicious versions - 1.82.7 and 1.82.8 - directly to the public registry. Any developer who updated LiteLLM during that window had malware running on their infrastructure that automatically harvested SSH keys, .env files, cloud provider credentials, cryptocurrency wallets, and AI API keys. Mercor was using LiteLLM in production. The company confirmed the breach on X, posting that it had \"moved promptly to contain and remediate the incident\" and was working with third-party forensics experts. But Lapsus$ had already listed Mercor on its dark web site as a live auction - \x3Ca href=\"https://cybersecuritynews.com/mercor-ai-data-breach/\" rel=\"noopener\" target=\"_blank\">\"Make an offer\"\x3C/a> - with the full alleged 4TB dataset. The scope of what Lapsus$ is claiming is significant: Mercor manages 100,000+ domain experts, including physicians, lawyers, and PhD researchers, and works directly with every major AI lab. Its databases could contain sensitive contractor information, identity documents from verification processes, and - per the leaked video files - recordings of AI model training sessions with expert contractors. As of publication, Mercor has not disclosed how many users were affected or confirmed the specific data Lapsus$ claims to hold. A class action law firm investigation has already been launched."},$R[679]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>This story is bigger than Mercor. Mercor is the first company to publicly confirm impact from the TeamPCP campaign - but SANS ISC's threat intelligence analysts are tracking AstraZeneca and Databricks as potential additional victims, with those organizations not yet acknowledging any breach. The LiteLLM attack vector is the nightmare scenario for the AI industry: a single compromised package with 97 million monthly downloads sitting at the exact point where every developer's API credentials flow. If you used LiteLLM versions 1.82.7 or 1.82.8 and haven't rotated credentials, SANS is explicitly warning that exploitation is actively underway. The broader context makes this week extraordinary: Anthropic accidentally exposed Mythos in a data misconfiguration, then accidentally exposed \x3Ca href=\"https://www.techmeme.com/260331/p69\" rel=\"noopener\" target=\"_blank\">500,000 lines of Claude Code source code\x3C/a> in a packaging error, and now Mercor - which trains Anthropic's own models - confirmed a major supply chain breach in the same seven-day window. The AI industry is moving fast and breaking things, and the things breaking this week are security perimeters. For founders: conduct an immediate audit of every open-source dependency in your production stack. The TeamPCP campaign specifically targeted developer credential stores - the exact infrastructure that most AI startups have deprioritized securing."},$R[680]={t:"h2",c:"\x3Cstrong>OPENAI CLOSES THE LARGEST PRIVATE FUNDING ROUND IN HISTORY - $122 BILLION AT AN $852 BILLION VALUATION\x3C/strong>"},$R[681]={t:"img",c:"/archive-images/620x-1-5483bc33.jpg"},$R[682]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>While the AI security world was on fire, OpenAI quietly closed the largest private funding round Silicon Valley has ever seen - \x3Ca href=\"https://www.cnbc.com/2026/03/31/openai-funding-round-ipo.html\" rel=\"noopener\" target=\"_blank\">$122 billion\x3C/a> in committed capital at an $852 billion post-money valuation. The company is generating $2 billion in revenue per month, serves more than 900 million weekly active users on ChatGPT, and is now describing itself not as an AI lab but as \"the core infrastru
1cture for AI.\" With a Q4 2026 IPO expected, this round is as much about anchoring public market expectations as it is about the capital itself."},$R[683]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>The round topped OpenAI's previously announced $110 billion in commitments, with the additional capital coming from a broader investor pool. SoftBank co-led alongside Andreessen Horowitz, D.E. Shaw Ventures, MGX, TPG, and T. Rowe Price Associates. Amazon ($50B), Nvidia ($30B), SoftBank ($30B), and Microsoft all participated. In an unprecedented move, $3 billion came from individual retail investors through bank channels - the first time OpenAI has opened a round to individuals - and the company will now be included in multiple ARK Invest ETFs, giving retail investors pre-IPO exposure without an S-1. The numbers behind the rise are striking: Codex now serves 2 million weekly users, up 5x in three months, with usage growing 70% month over month. Enterprise revenue has climbed from 30% to 40% of total revenue and is on track to reach parity with consumer by year-end. OpenAI's APIs now process more than 15 billion tokens per minute - meaning the platform processes the equivalent of every book ever printed in human history roughly every 16 hours. The company is simultaneously planning a \x3Ca href=\"https://techcrunch.com/2026/03/31/openai-not-yet-public-raises-3b-from-retail-investors-in-monster-122b-fund-raise/\" rel=\"noopener\" target=\"_blank\">\"superapp\"\x3C/a> combining ChatGPT, Codex, and agentic capabilities, and has already shut down the consumer-focused Sora video app to redirect resources toward enterprise productivity tools. Amazon's $50 billion investment includes a key condition: it's contingent on OpenAI either achieving AGI or completing an IPO - a clause that structurally accelerates the public market timeline regardless of Sam Altman's preferences."},$R[684]={t:"p",c:"Why it matters: At $852 billion, OpenAI is now valued higher than every public company except Apple, Nvidia, Microsoft, Alphabet, Amazon, Meta, and Berkshire Hathaway - without being public, without being profitable, and without a clear path to profitability until 2030. The round's structure tells you where the company is headed: ARK ETF inclusion and retail investor access are not things you do when you plan to stay private for years. This is IPO preparation dressed as a funding round. For founders and investors, the most important number is not the valuation - it's the 40% enterprise revenue share growing toward 50% . That's the signal that OpenAI is successfully transitioning from a consumer chatbot to enterprise infrastructure, the only business model that can justify a $1 trillion public market valuation. The risk is real: OpenAI is burning cash at record rates toward a 2030 profitability target, in a market where Anthropic, Google, and open-source competitors are closing the capability gap. The question isn't whether OpenAI is impressive - it clearly is. The question is whether the moat is durable enough to earn the multiple. Watch the IPO filing for the answer."},$R[685]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.00.12_PM-ef59bfe3.png"},$R[686]={t:"h2",c:"\x3Cstrong>Global AI Quick Hits\x3C/strong>"},$R[687]={t:"p",c:"\x3Ca href=\"https://www.npr.org/2026/03/24/nx-s1-5759276/anthropic-pentagon-claude-preliminary-injunction-hearing\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>Federal judge rules in favor of Anthropic - preliminary injunction granted\x3C/strong>\x3C/a>\x3Ca href=\"https://www.npr.org/2026/03/24/nx-s1-5759276/anthropic-pentagon-claude-preliminary-injunction-hearing\" rel=\"noopener\" target=\"_blank\"> \x3C/a>- Judge Rita F. Lin sided with Anthropic, pausing the Pentagon's supply-chain risk designation while the full lawsuit proceeds. The ruling calls out the government's conduct as likely unconstitutional retaliation and gives Anthropic breathing room to fight from a position of relative strength. Watch the full case closely - the ruling will set the precedent for every AI company's government contract negotiations going forward."},$R[688]={t:"p",c:"\x3Ca href=\"https://llm-stats.com/ai-news\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>Global VC hits a record $297 billion in Q1 2026 - AI captures 81% of it\x3C/strong>\x3C/a> - Crunchbase reported that Q1 2026 set a record for global venture investment, with AI startups capturing more than four out of every five dollars deployed. Just four companies raised 64% of the total. The AI funding boom is not slowing - it's concentrating."},$R[689]={t:"p",c:"\x3Ca href=\"https://www.techmeme.com/260331/p69\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>Anthropic issues 8,000+ copyright takedowns after Claude Code source leak\x3C/strong>\x3C/a> - After the Claude Code source code spread across GitHub and developer forums, Anthropic's legal team began issuing DMCA takedown requests to remove copies. Developers noted the leaked code contains anti-distillation systems Anthropic built into Claude Code - including one that injects fake tool calls into sessions to corrupt competitor training data."},$R[690]={t:"p",c:"\x3Ca href=\"https://llm-stats.com/ai-news\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>Perplexity faces a class action lawsuit for allegedly sharing user data with Meta and Google\x3C/strong>\x3C/a> - Bloomberg reported that Perplexity is being sued for allegedly sharing personal user data from chat sessions with Meta and Google. The case is the latest sign that AI companies' data practices are entering a new phase of legal scrutiny - and that the \"privacy by default\" positioning many AI tools claim may not hold up under examination."},$R[691]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.01.58_PM-b002921b.png"},$R[692]={t:"h2",c:"\x3Cstrong>\x3Ca href=\"https://www.skool.com/ai-insider/about\" rel=\"noopener\" target=\"_blank\">Join Our AI Community\x3C/a>\x3C/strong>"},$R[693]={t:"p",c:"Most people are reading about AI. AI Insider members are using it to work faster and stay ahead. Every week I go live and break down one AI tool or workflow - exactly how I'm using it, what's working, what's not. You ask questions, I answer live, and everything gets saved to the vault. Join today, and 
1you'll unlock 20+ past sessions instantly, plus weekly live breakdowns, an AI tools database, step-by-step courses, early access to new AI companies, and a community of 300+ founders and builders actually putting this stuff to work. Total value is over $7,400 - click below to unlock the vault today."},$R[694]={t:"p",c:"- Drew & the rest of the humans behind The AI Debrief"}],edition:"debrief",issueNo:null},$R[695]={slug:"the-lobster-just-changed-everything",title:"The Lobster Just Changed Everything",subtitle:"Jensen Huang declared AGI. OpenClaw took over China. And a federal judge just sided with Anthropic.",date:"2026-03-25",thumbnail:"/archive-images/-1x-1__1_-425385e2.webp",blocks:$R[696]=[$R[697]={t:"img",c:"/archive-images/debrief_logo_3-5be9874b.png"},$R[698]={t:"p",c:"This week, the story everyone in AI has been circling for years finally got said out loud. Jensen Huang told Lex Fridman that AGI has been achieved - and his example wasn't some theoretical supercomputer. It was OpenClaw, the open-source AI agent that went from zero to 250,000 GitHub stars in under six months and is now being installed on laptops in Beijing by the hundreds, integrated into WeChat, and used by millions of people to find jobs, run errands, and build apps autonomously."},$R[699]={t:"p",c:"Meanwhile, a federal judge told the Pentagon its Anthropic ban looks like punishment - and said she expects to rule within days."},$R[700]={t:"p",c:"Here's the full breakdown."},$R[701]={t:"p",c:"\x3Cstrong>In today’s AI debrief:\x3C/strong>"},$R[702]={t:"ul",c:$R[703]=["Jensen Huang declares AGI is here - and points to OpenClaw as exhibit A","OpenClaw goes fully global: 250K GitHub stars, WeChat integration, China government crackdowns, and a major v2026.3.22 update","Federal Judge Lin calls the Anthropic ban \"an attempt to cripple\" the company - ruling expected within days","Quick hits on OpenAI's $10B raise, Claude's computer use lau
1nch, Tencent's WeChat integration, and the CFO survey that should scare every white-collar worker"]},$R[704]={t:"h2",c:"\x3Cstrong>\x3Ca href=\"https://www.skool.com/ai-insider/about\" rel=\"noopener\" target=\"_blank\">Join Our AI Community\x3C/a>\x3C/strong>"},$R[705]={t:"p",c:"Most people are reading about AI. AI Insider members are using it to work faster and stay ahead. Every week I go live and break down one AI tool or workflow - exactly how I'm using it, what's working, what's not. You ask questions, I answer live, and everything gets saved to the vault. Join today, and you'll unlock 20+ past sessions instantly, plus weekly live breakdowns, an AI tools database, step-by-step courses, early access to new AI companies, and a community of 300+ founders and builders actually putting this stuff to work. Total value is over $7,400 - click below to unlock the vault today."},$R[706]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_5.51.07_PM-fab03707.png"},$R[707]={t:"h2",c:"\x3Cstrong>\x3Ca href=\"https://www.ibtimes.com/nvidia-ceo-jensen-huang-declares-we-have-achieved-agi-sparking-debate-over-definition-3800074\" rel=\"noopener\" target=\"_blank\">JENSEN HUANG SAYS \"I THINK WE'VE ACHIEVED AGI\" - AND OPENCLAW IS WHY\x3C/a>\x3C/strong>"},$R[708]={t:"img",c:"/archive-images/STKP210_JENSEN_HUANG_B-f47e785c.webp"},$R[709]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>On the Lex Fridman podcast released March 23, Nvidia CEO Jensen Huang made the most attention-grabbing four words in AI this year: \"I think we've achieved AGI.\" His definition wasn't the sci-fi version - it was Fridman's specific framing of whether AI can start, grow, and run a company worth over $1 billion. Huang said yes, and his primary example was OpenClaw. The clip went viral within hours, hitting \x3Ca href=\"https://finance.yahoo.com/news/nvidia-ceo-jensen-huang-claims-agi-has-been-achieved-can-create-billion-dollar-businesses-172225126.html\" rel=\"noopener\" target=\"_blank\">4.7 million views\x3C/a> on Polymarket's post alone, and reignited the debate about whether the industry is genuinely at a new threshold - or whether it's redefining the goalpost to fit where it already is."},$R[710]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>Huang's exact framing is important to understand before you decide what to make of it. Fridman asked whether an AI could autonomously build a billion-dollar company - five, ten, or twenty years from now. Huang replied without hesitation: \x3Ca href=\"https://lexfridman.com/jensen-huang-transcript/\" rel=\"noopener\" target=\"_blank\">\"I think it's now.\"\x3C/a> He immediately hedged: \"You said a billion, and you didn't say forever.\" He argues that OpenClaw agents are already being used to launch social applications, create viral digital content, and run autonomous workflows that generate real economic value - briefly, at scale, without constant human direction. He envisioned an AI creating \"some interesting little app that all of a sudden a few billion people used for 50 cents\" and then disappearing - which, he argued, matches the bar Fridman set. He also pulled back sharply when pressed further, admitting: \"The odds of 100,000 of those agents building Nvidia is zero percent.\" The reaction split the industry. Supporters called it a pragmatic acknowledgment of where agentic AI actually is in early 2026. Critics said Huang conveniently moved the goalposts on a term that has enormous contractual and regulatory significance - pointing out that OpenAI's charter defines AGI as systems that \"outperform humans at most economically valuable work,\" a standard vastly beyond a viral app that \"dies away.\" The Lex Fridman transcript reveals Huang spent significant time on OpenClaw's architecture - calling it \"what ChatGPT did for generative systems,\" and describing how it reinvented the computer by giving AI access to files, the web, APIs, and code execution in one integrated loop."},$R[711]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>Whether or not Huang is right about AGI, the statement he made has real consequences. When the CEO of the company powering 80% of AI training declares AGI achieved, it shifts capital allocation, regulatory timelines, and hiring plans at every major enterprise. The subtext is deliberate: if AGI is here, the demand for Nvidia's chips has no near-term ceiling. For founders, the more useful takeaway is the OpenClaw framing - Huang is telling you that the benchmark for AI value creation has moved from \"can it answer questions?\" to \x3Ca href=\"https://www.thestreet.com/technology/nvidia-ceo-je
1nsen-huang-says-we-have-achieved-agi\" rel=\"noopener\" target=\"_blank\">\"can it autonomously build something real?\"\x3C/a> The companies that survive the next two years will be the ones designing for autonomous execution, not just smarter responses. And Huang's hedge matters too: short-lived, narrow economic value is not the same as durable business moats. Don't confuse a viral agent app with a defensible company."},$R[712]={t:"h2",c:"\x3Cstrong>\x3Ca href=\"https://www.cnbc.com/2026/03/21/openclaw-chatgpt-moment-sparks-concern-ai-models-becoming-commodities.html\" rel=\"noopener\" target=\"_blank\">THE LOBSTER TAKES OVER: OPENCLAW HITS 250K GITHUB STARS, ENTERS WECHAT, AND TRIGGERS GOVERNMENT CRACKDOWNS IN CHINA\x3C/a>\x3C/strong>"},$R[713]={t:"img",c:"/archive-images/c5d90d1d806889d8906da0ca16e9eab19ad5b9c3-20682ed0.webp"},$R[714]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>OpenClaw - the open-source AI agent that Jensen Huang called the \"operating system for personal AI\" at GTC - had its biggest week yet. Tencent integrated it directly into WeChat, giving over one billion monthly active users access to AI agents via chat. The project hit 250,000 GitHub stars. Version 2026.3.22 dropped with a ClawHub marketplace, multi-model sub-agents, and critical session fixes. And China's government issued its first formal security warnings about the tool, with nearly \x3Ca href=\"https://www.cnbc.com/2026/03/12/china-openclaw-ai-agent-adoption-tech-companies-government-support-lobster-shrimp.html\" rel=\"noopener\" target=\"_blank\">23,000 users'\x3C/a> assets exposed to cyberattack. The lobster, as Chinese users call it, is everywhere - and the world is just now starting to reckon with what that means."},$R[715]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>OpenClaw's China moment is one of the fastest grassroots technology adoptions in recent memory. Tencent held free in-person OpenClaw setup sessions in Shenzhen, helping hundreds of users install the tool on TencentCloud. ByteDance's cloud unit launched ArkClaw, a browser-based version that eliminates local setup. JD.com and Meituan partnered with Lenovo to offer paid remote installation services. Baidu engineers are photographed installing it on laptops at their Beijing headquarters. The tool is being used by young Chinese job seekers to autonomously scan listings, apply, prep for interviews, and track applications - with one 24-year-old Shanghai user telling NBC News it saves him three hours per day. But it's not all upside: China's National Cybersecurity Alert Center warned that 23,000 OpenClaw users had their assets exposed to the internet and are \x3Ca href=\"https://www.pymnts.com/artificial-intelligence-2/2026/tencent-adds-openclaw-ai-agent-to-chinas-most-popular-app/\" rel=\"noopener\" target=\"_blank\">\"highly likely to become priority targets for cyberattack.\" \x3C/a>The Chinese Academy of Information and Communications Technology is now developing formal standards for \"claw\" agents, and state-owned enterprises, universities, and government employees have begun restricting or banning it - mirroring similar restrictions at U.S. companies. On the product side, OpenClaw v2026.3.22 dropped this week with ClawHub marketplace integration (think an app store for agents), the /btw command for side conversations mid-task, adjustable sub-agent thinking levels to control costs, and multi-model sub-agents - letting cheaper models handle simple tasks while reserving expensive frontier models for complex reasoning. The economics matter: with Claude Opus 4.6's 1-million-token context window, cost-unaware OpenClaw deployments can burn through API budgets fast."},$R[716]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>CNBC called OpenClaw's moment \"AI's ChatGPT moment\" - the point where a technology stops being a developer curiosity and becomes a consumer phenomenon. The difference is that, unlike ChatGPT, OpenClaw is open-source, runs locally, and works with any model, which means OpenAI, Anthropic, and Google don't own it and can't control it. One industry CEO called it \x3Ca href=\"https://nvidianews.nvidia.com/news/nvidia-announces-nemoclaw\" rel=\"noopener\" target=\"_blank\">\"the black swan moment most big AI companies feared.\" \x3C/a>For founders: the ClawHub marketplace is the first serious attempt to build an app store layer on top of agentic AI, which means distribution, monetization, and platform dynamics are about to emerge in the agent space the same way they emerged on iOS and Android. The security risks are real and will only grow as more enterprise data flows through autonomous agents. If you're building on OpenClaw or deploying it internally, read the v2026.3.22 session management fixes - the bug where stale cron job sessions accumulate silently is affecting production deployments right now."},$R[717]={t:"h2",c:"\x3Cstrong>\x3Ca href=\"https://www.npr.org/2026/03/24/nx-s1-5759276/anthropic-
1pentagon-claude-preliminary-injunction-hearing\" rel=\"noopener\" target=\"_blank\">FEDERAL JUDGE SAYS ANTHROPIC BAN LOOKS LIKE \"AN ATTEMPT TO CRIPPLE\" THE COMPANY - RULING IMMINENT\x3C/a>\x3C/strong>"},$R[718]={t:"img",c:"/archive-images/69c34a1434f2d.image-c2364729.jpg"},$R[719]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>The March 24 hearing everyone was watching delivered exactly the signal Anthropic needed. U.S. District Judge Rita F. Lin opened Tuesday's preliminary injunction hearing by telling the courtroom that the Pentagon's blacklisting of Anthropic \"looks like an attempt to cripple Anthropic\" and that the government's conduct was \x3Ca href=\"https://www.cnbc.com/2026/03/24/anthropic-lawsuit-pentagon-supply-chain-risk-claude.html\" rel=\"noopener\" target=\"_blank\">\"troubling,\"\x3C/a> specifically questioning whether it constituted illegal punishment for speech. She said she expects to issue a ruling within days. If granted, the preliminary injunction would pause the supply-chain risk designation while the full lawsuit proceeds. If denied, Anthropic continues fighting the blacklist while it actively erodes enterprise relationships."},$R[720]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>Judge Lin's remarks were the starkest judicial signal yet that the Trump administration's legal theory may not hold up. She pressed the government's lawyer on what would justify a supply-chain risk designation under the relevant statute, pushing back on the DOD's argument that Anthropic's mere stubbornness in contract negotiations could qualify a company for a designation historically reserved for foreign adversaries like Huawei. \"What I'm hearing from you is that it's enough if an IT vendor is stubborn and insists on certain terms and asks annoying questions, then it can be designated as a supply chain risk,\" Lin said. \x3Ca href=\"https://www.ibtimes.co.uk/federal-judge-calls-hegseth-directive-troubling-anthropic-case-suggests-it-may-cripple-company-1788238\" rel=\"noopener\" target=\"_blank\">\"That seems a pretty low bar.\" \x3C/a>She separately questioned whether the government violated the law not just by canceling the contract - which she said the Pentagon has every right to do - but by extending the ban to all commercial activity with any Pentagon contractor or partner. Anthropic's lawyer Michael Mongan made clear the ask is narrow: the injunction would not force the government to use Claude, only prevent the extra-contractual punishment. Breaking Defense reported that Trump and Hegseth's own public social media posts may have undermined the government's legal position, with one attorney calling Trump's Truth Social posts \"admissions against interest\" that signal retaliatory intent. A ruling is expected before the end of the week."},$R[721]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>A preliminary injunction would be a significant win for Anthropic - not because it resolves the underlying case, but because it stops the hemorrhaging. \x3Ca href=\"https://www.aljazeera.com/economy/2026/3/24/anthropic-challenges-us-pentagons-ban-in-san-francisco-court-showdown\" rel=\"noopener\" target=\"_blank\">Over 100 enterprise customers\x3C/a> have already contacted Anthropic with concerns, and the company's CFO has told the court the designation could cost multiple billions in 2026 revenue. Every day the designation stays live, more enterprise procurement teams quietly route away from Claude out of caution. For the broader AI industry, Judge Lin's framing is the most important signal in this case so far: she's treating this as a question of government overreach and speech retaliation, not a question of national security deference. That framing makes it significantly harder for the government to win on the First Amendment claim. Watch for the ruling - it will reshape how every AI company thinks about government contract negotiations for years."},$R[722]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.00.12_PM-ef59bfe3.png"},$R[723]={t:"h2",c:"\x3Cstrong>Global AI Quick Hits\x3C/strong>"},$R[724]={t:"p",c:"\x3Cstrong>\x3Ca href=\"https://www.cryptointegrat.com/p/ai-news-march-25-2026\" rel=\"noopener\" target=\"_blank\">OpenAI raises another $10 billion at $730B valuation - with a 17.5% guaranteed return \x3C/
1a>\x3C/strong>- OpenAI is seeking a fresh $10B round from private equity firms, reportedly offering investors a guaranteed minimum return of 17.5% to attract capital. The terms signal that even at $730 billion, OpenAI needs creative financing structures to keep pace with $600 billion in projected infrastructure spending through 2030."},$R[725]={t:"p",c:"\x3Cstrong>\x3Ca href=\"https://www.theneuron.ai/ai-news-digests/around-the-horn-digest-everything-that-happened-in-ai-on-wednesday-march-25-2026-/\" rel=\"noopener\" target=\"_blank\">Claude gets computer use - opens your apps, fills your spreadsheets, browses your browser\x3C/a>\x3C/strong> - Anthropic launched computer use for Claude in research preview via Cowork and Claude Code on macOS. Claude can now take over your screen, navigate browsers, fill spreadsheets, and complete tasks the way you would sitting at your desk. This is the most significant capability expansion since Claude Code launched."},$R[726]={t:"p",c:"\x3Cstrong>\x3Ca href=\"https://techstartups.com/2026/03/25/top-tech-news-today-march-25-2026/\" rel=\"noopener\" target=\"_blank\">CFO survey: AI is now in workforce budgets, not just roadmaps\x3C/a>\x3C/strong> - A Wall Street Journal survey of chief financial officers found that companies are now formally planning headcount reductions driven by AI automation, with administrative roles most exposed. This is the transition from \"AI might affect jobs\" to \"AI is already in our 2026 staffing model.\""},$R[727]={t:"p",c:"\x3Cstrong>\x3Ca href=\"https://techstartups.com/2026/03/24/top-tech-news-today-march-24-2026/\" rel=\"noopener\" target=\"_blank\">OpenAI to nearly double headcount to 8,000 by year-end\x3C/a>\x3C/strong> - Even as companies cut workers citing AI, OpenAI is hiring aggressively across product, engineering, sales, and enterprise - a sign that frontier AI is evolving from research lab to full-scale operating company. The model race still matters, but so does the sales force behind it."},$R[728]={t:"p",c:"- Drew & the rest of the humans behind The AI Debrief"}],edition:"debrief",issueNo:null},$R[729]={slug:"1-trillion-90-days-and-the-ai-trial-of-the-century",title:"$1 Trillion. 90 Days. And the AI Trial of the Century.",subtitle:"Nvidia doubled its forecast. Morgan Stanley issued a warning. Anthropic heads to court.",date:"2026-03-19",thumbnail:"/archive-images/JensenHuang_GTC-feaa4998.webp",blocks:$R[730]=[$R[731]={t:"p",c:"\x3Cstrong>In partnership with\x3C/strong>"},$R[732]={t:"img",c:"/archive-images/HubSpotlogo-8c869a59.png"},$R[733]={t:"h3",c:"Want to get the most out of ChatGPT?"},$R[734]={t:"img",c:"/archive-images/CNN_Creative_Refresh_4B-36adad5b.jpg"},$R[735]={t:"p",c:"ChatGPT is a superpower if you know how to use it correctly."},$R[736]={t:"p",c:"Discover how \x3Ca href=\"https://offers.hubspot.com/using-chatgpt-at-work?_bhiiv=opp_e728eb89-da5e-4356-9eb3-93ac048e282e_b942af4d&amp;bhcl_id=077432d0-5921-4c64-888f-845ba9e3931a_%7B%7Bsubscriber_id%7D%7D_%7B%7Bemail_address_id%7D%7D\" rel=\"noopener\" target=\"_blank\">HubSpot's guide to AI\x3C/a> can elevate both your productivity and creativity to get more things done."},$R[737]={t:"p",c:"Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI."},$R[738]={t:"p",c:"\x3Ca href=\"https://offers.hubspot.com/using-chatgpt-at-work?_bhiiv=opp_e728eb89-da5e-4356-9eb3-93ac048e282e_b942af4d&amp;bhcl_id=077432d0-5921-4c64-888f-845ba9e3931a_%7B%7Bsubscriber_id%7D%7D_%7B%7Bemail_address_id%7D%7D\" rel=\"noopener\" target=\"_blank\">Download the free guide\x3C/a>"},$R[739]={t:"img",c:"/archive-images/debrief_logo_3-80b279c1.png"},$R[740]={t:"p",c:"This week had three stories that, taken together, tell you everything you need to know about where the AI industry is headed. Nvidia just doubled its revenue forecast to $1 trillion. Morgan Stanley is warning that a capability breakthrough is arriving in 90 days and most of the world isn't ready. And the Anthropic lawsuit - already the most consequential AI legal case in history - just got a massive new twist."},$R[741]={t:"p",c:"Buckle up. Here's what you missed."},$R[742]={t:"p",c:"\x3Cstrong>In today’s AI debrief:\x3C/strong>"},$R[743]={t:"ul",c:$R[744]=["Jensen Huang drops a $1 trillion demand forecast at GTC 2026 - and announces AI data centers in space","Morgan Stanley warns a \"massive AI breakthrough\" hits in April–June 2026, and executives at top AI labs say it will \"shock\" investors","The DOJ files its 40-page rebuttal calling Anthropic an \"unacceptable risk to national security\" - 149 former federal judges file back in support","Quick hits on OpenAI's Q4 IPO push, Meta's $27B infrastru
1cture deal, Mistral Small 4, and ServiceNow's brutal jobs forecast"]},$R[745]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_5.51.07_PM-fab03707.png"},$R[746]={t:"h2",c:"\x3Cstrong>JENSEN HUANG RAISES AI DEMAND FORECAST TO $1 TRILLION - AND ANNOUNCES DATA CENTERS IN SPACE\x3C/strong>"},$R[747]={t:"img",c:"/archive-images/7034723e61da7abda89f2dad978a14d0-26229245.jpeg"},$R[748]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>At his sold-out GTC 2026 keynote in San Jose, Nvidia CEO Jensen Huang announced the Vera Rubin AI platform, raised his demand forecast from $500 billion to $1 trillion through 2027, unveiled the Groq 3 LPU inference chip, launched NemoClaw as the \x3Ca href=\"https://blogs.nvidia.com/blog/gtc-2026-news/\" rel=\"noopener\" target=\"_blank\">\"operating system for agentic AI,\"\x3C/a> and - in perhaps the most unexpected moment of the night - announced that Nvidia is taking its AI data centers to orbit with Vera Rubin Space-1. It was two and a half hours of announcements that made every other company's product roadmap look small."},$R[749]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>Vera Rubin is Nvidia's most integrated system ever - seven chip types, five rack-scale computers, operating as a single AI supercomputer. The headline number: 3.6 exaflops of compute and 260 terabytes per second of all-to-all bandwidth. Third-party analysis from Semi Analysis found Vera Rubin delivers roughly 50x more tokens per watt than the prior-generation Hopper H200. Combined with the newly unveiled Groq 3 LPU rack - the product of Nvidia's $20 billion Groq acquisition - the full system delivers 35x more throughput per watt than the previous Blackwell generation and opens what Huang called a \"$300 billion annual revenue opportunity\" from the inference layer alone. The $1 trillion demand forecast - doubled from Huang's $500 billion figure from just a year ago - was framed as a conservative estimate, with Huang suggesting actual demand could outpace supply. He described computing demand as having jumped 10,000 times over the last two years. NemoClaw, Nvidia's open-source agentic AI platform built in partnership with OpenClaw, was framed as the \x3Ca href=\"https://www.eweek.com/news/nvidia-gtc-2026-keynote-jensen-huang/\" rel=\"noopener\" target=\"_blank\">\"Linux of the agentic era.\"\x3C/a> And the space announcement - Vera Rubin Space-1, designed to power orbital data centers - drew laughs and gasps simultaneously. Huang acknowledged the obvious challenge: \"In space, there's no conduction. There's no convection. There's just radiation, so we have to figure out how to cool these systems.\" Disney's free-roaming Olaf robot joined him onstage, powered by Nvidia's AI and deep reinforcement learning - a preview of its Disneyland Paris debut on March 29."},$R[750]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>Nvidia just told the world that the AI infrastructure boom is not slowing down - it's doubling. The $1 trillion demand signal through 2027 is the single most important data point in AI investing right now, because it tells you that every hyperscaler's CapEx guidance is going to stay elevated or go higher. The inference story is what every founder should focus on: training built the boom, but inference is where ongoing revenue gets made. As AI models run longer, reason deeper, and act more like software agents than chatbots, the economics of tokens per watt become the margin story of the decade. For VC-backed startups: the chip access problem is getting solved, but the cost curve is moving in both directions. More computing availability means more competition. The companies that survive won't just have access to Vera Rubin - they'll have built applications that can justify the \x3Ca href=\"https://www.datacenterknowledge.com/data-center-chips/gtc-2026-nvidia-unveils-vera-rubin-ai-platform-eyes-1t-by-2027\" rel=\"noopener\" target=\"_blank\">premium token price Nvidia is targeting.\x3C/a> Space data centers may sound like a joke, but orbital solar power and zero-atmosphere cooling could one day make it the cheapest computing on earth. Take note."},$R[751]={t:"h2",c:"\x3Cstrong>MORGAN STANLEY: A \"MASSIVE AI BREAKTHROUGH\" IS 90 DAYS AWAY - AND MOST OF THE WORLD ISN'T READY\x3C/strong>"},$R[752]={t:"img",c:"/archive-images/morgan-stanley-warns-an-ai-breakthrough-is-coming-in-2026-v0-803f66d4.webp"},$R[753]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>In a sweeping new research report, Morgan Stanley warned that a transformative leap in AI capability is likely to hit in April–June 2026, driven by an unprecedented accumulation of compute at America's top AI labs. Executives at major AI labs are privately telling investors to brace for progress that will \"shock\" them. OpenAI's GPT-5.4 \"Thinking\" model already scored 83% on the GDPVal benchmark - at or above human expert level on economically valuable tasks - and Morgan Stanley says the market \"is not prepared for the non-linear increase in LLM capabilities\" that is about to materialize."},$R[754]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>Morgan Stanley's thesis is built on scaling laws that, according to the report, are still holding firm. The bank cites Elon Musk's argument that applying 10x the compute to LLM training effectively doubles a model's intelligence - and says the data backs that claim. With AI labs now running some of the largest compute cluster
1s in history, a capability step-change is the mathematical conclusion. The economic consequences the bank forecasts are severe: AI becomes a \x3Ca href=\"https://finance.yahoo.com/news/morgan-stanley-warns-ai-breakthrough-072000084.html\" rel=\"noopener\" target=\"_blank\">\"deflationary force\"\x3C/a> as systems replicate human work at a fraction of the cost, executives execute large-scale workforce reductions, and Sam Altman's vision of one-to-five-person companies outcompeting large incumbents starts becoming real. The infrastructure demands are equally staggering - Morgan Stanley estimates the U.S. could face a power shortage of 9 to 18 gigawatts by 2028 as AI data centers consume electricity at city scale, with developers already converting Bitcoin mining facilities and installing natural gas turbines directly at data centers to keep up. The most alarming footnote: xAI co-founder Jimmy Ba has suggested that recursive self-improvement loops - where AI systems autonomously upgrade their own capabilities - could emerge as early as the first half of 2027."},$R[755]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>When Morgan Stanley says a capability jump is 90 days out and executives at top labs are calling it \"shocking,\" that is no longer a research curiosity - it's a signal that institutional capital is now pricing in a discontinuous leap. For founders: the question is no longer \"when will AI be good enough to automate X?\" It's \"are we building something that survives when the next capability threshold lands?\" The recursive self-improvement timeline is the most provocative detail - if AI begins meaningfully accelerating its own development in 2027, every product roadmap built beyond that point is speculative. The power crisis is also real: the 9-18 gigawatt shortfall Morgan Stanley projects by 2028 is larger than the entire electricity consumption of several U.S. states. Energy infrastructure is the next AI bottleneck - and the companies that solve it first will have a moat that no model can replicate."},$R[756]={t:"h2",c:"\x3Cstrong>DOJ CALLS ANTHROPIC AN \"UNACCEPTABLE RISK\" - 149 FORMER FEDERAL JUDGES FILE BACK IN SUPPORT\x3C/strong>"},$R[757]={t:"img",c:"/archive-images/buisness_DOD_Anthropic_GettyImages-2265262796-b7d8f3aa.webp"},$R[758]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>The Anthropic-Pentagon legal battle escalated dramatically this week. The DOJ filed a 40-page rebuttal calling Anthropic an \x3Ca href=\"https://www.aljazeera.com/economy/2026/3/18/trump-administration-defends-anthropic-blacklisting-in-us-court\" rel=\"noopener\" target=\"_blank\">\"unacceptable risk to national security,\"\x3C/a> arguing the company could disable or secretly alter Claude during active military operations. In the same 24 hours, a bipartisan coalition of 149 former federal and state judges filed an amicus brief backing Anthropic, calling the supply-chain designation \"substantively and procedurally unlawful.\" A preliminary injunction hearing is set for March 24. That date will determine whether Anthropic walks into court with the designation paused - or fighting from behind."},$R[759]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>The DOJ's 40-page filing is the government's most aggressive legal response yet. Attorneys argued that Anthropic's safety \"red lines\" - its refusals to allow Claude in autonomous weapons or domestic surveillance - are not protected speech but rather conduct, and that the company's ongoing role as developer and maintainer of Claude creates inherent risk: they could \x3Ca href=\"https://www.axios.com/2026/03/19/pentagon-anthropic-foreign-workforce-security-risks\" rel=\"noopener\" target=\"_blank\">\"subvert the design and/or functionality\"\x3C/a> of their tools or preemptively alter model behavior during warfighting. The Pentagon doubled down separately, with Under Secretary Emil Michael filing a declaration highlighting a new national security concern: Anthropic employs \"a large number of foreign nationals to build and support its LLM products, including many from the People's Republic of China.\" The 149 former federal judges - filing on a bipartisan basis - countered that the DOD had misread the statute entirely, arguing that supply-chain risk law \"narrowly limits\" its application to malicious conduct like sabotage by hostile actors, not policy disagreements with domestic companies. Anthropic's CFO has already told the court that the designation could cost the company multiple billions in 2026 revenue, with over 100 enterprise customers making inquiries about their exposure. Microsoft, Amazon, Apple, and Google have all c
1onfirmed they will continue offering Claude through their platforms for non-Pentagon work."},$R[760]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>The March 24 hearing is the most important date in AI law this year. If a federal judge grants Anthropic's preliminary injunction, the designation is paused while the case proceeds - and Anthropic buys time to fight from a position of relative strength. If the injunction is denied, the designation stays live while appeals play out and enterprise customer attrition accelerates. The 149 judges’ amicus brief signals to the court that legal experts across party lines view the government's statutory interpretation as a serious overreach. The foreign nationals’ argument is the most politically charged new development: it signals the government is willing to use Anthropic's global workforce as a national security lever even though, as the \x3Ca href=\"https://www.washingtonexaminer.com/policy/defense/4495348/pentagon-files-first-rebuttal-anthropic-lawsuit-unacceptable-risk/\" rel=\"noopener\" target=\"_blank\">Foundation for American Innovation\x3C/a> noted, Anthropic is widely considered the most proactive company in the AI industry at policing insider threats. Watch March 24."},$R[761]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.00.12_PM-ef59bfe3.png"},$R[762]={t:"h2",c:"\x3Cstrong>Global AI Quick Hits\x3C/strong>"},$R[763]={t:"p",c:"\x3Ca href=\"https://www.cnbc.com/2026/03/17/openai-preps-for-ipo-in-2026-says-chatgpt-must-be-productivity-tool.html\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>OpenAI preps Q4 2026 IPO at a potential $1 trillion valuation\x3C/strong>\x3C/a> - CNBC confirmed OpenAI is targeting a Q4 2026 public listing. The company has hit $25 billion in annualized revenue - growing 17% from year-end 2025 - and is orienting aggressively toward enterprise productivity tools. An IPO at $1 trillion would be the largest public offering in history, surpassing Saudi Aramco's $25.6 billion raise in 2019. The catch: OpenAI doesn't expect to turn a profit until 2030 and faces a projected $207 billion funding shortfall by then."},$R[764]={t:"p",c:"\x3Ca href=\"https://techstartups.com/2026/03/17/top-tech-news-today-march-17-2026/\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>Meta locks in $27 billion AI infrastructure deal with Nebius\x3C/strong>\x3C/a> - Meta signed a five-year agreement giving it large-scale data center capacity from specialist cloud provider Nebius, underscoring how hyperscale AI demand is flowing to a new class of infrastructure intermediaries - not just traditional public cloud giants."},$R[765]={t:"p",c:"\x3Ca href=\"https://radicaldatascience.wordpress.com/2026/03/19/ai-news-briefs-bulletin-board-for-march-2026/\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>Mistral Small 4 launches as an open-source multimodal powerhouse\x3C/strong>\x3C/a> - Mistral dropped Small 4 this week: 119 billion parameters, Mixture of Experts architecture, unified text and image inputs, configurable reasoning effort, and full open-source access on vLLM, llama.cpp, and HuggingFace. It's being positioned as a single model that replaces the Magistral, Pixtral, and Devstral lineup entirely."},$R[766]={t:"p",c:"\x3Ca href=\"https://aiagentstore.ai/ai-agent-news/this-week\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>ServiceNow CEO warns college grad unemployment could hit 30% in two years\x3C/strong>\x3C/a>\x3Ca href=\"https://aiagentstore.ai/ai-agent-news/this-week\" rel=\"noopener\" target=\"_blank\"> \x3C/a>- ServiceNow's CEO said AI agents are taking over entry-level tasks fast enough that unemployment among recent graduates could reach 30% by 2028. Current data already shows 5.7% unemployment and 42.5% underemployment for recent college graduates - a cohort entering the worst white-collar job market in decades precisely as agentic AI matures."},$R[767]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.01.58_PM-b002921b.png"},$R[768]={t:"p",c:"- Drew & the rest of the humans behind The AI Debrief"}],edition:"toolkit",issueNo:null},$R[769]={slug:"anthropic-sues-the-pentagon",title:"Anthropic Sues the Pentagon",subtitle:"#QuitGPT hits 2.5 million. Claude hits #1. Oracle fires 30,000.",date:"2026-03-13",thumbnail:"/archive-images/p-1-91505493-anthropic-sues-pentagon-fa3d4a46.webp",blocks:$R[770]=[$R[771]={t:"img",c:"/archive-images/debrief_logo_3-b31ced3c.png"},$R[772]={t:"p",c:"Last week, we told you to watch at 5:01 PM on Frid
1ay. Here's what happened: Anthropic didn't blink. The Pentagon labeled them a national security risk. Anthropic filed two federal lawsuits. And 2.5 million ChatGPT users rage-quit in protest of OpenAI taking the deal Anthropic walked away from."},$R[773]={t:"p",c:"We are watching the AI industry split into two camps in real time - and this week made the fault lines impossible to ignore."},$R[774]={t:"p",c:"\x3Cstrong>In today’s AI debrief:\x3C/strong>"},$R[775]={t:"ul",c:$R[776]=["Anthropic sues the Pentagon after being labeled a supply-chain risk, putting $60B in investor capital on the line","#QuitGPT explodes to 2.5 million users - Claude hits #1 on the App Store for the first time","Oracle and Block announce 34,000 layoffs in a single week, both explicitly crediting AI","Quick hits on Nvidia's GTC \"Super Bowl,\" the Senate approving AI for staff, GPT-5.4, and more"]},$R[777]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_5.51.07_PM-fab03707.png"},$R[778]={t:"h2",c:"\x3Cstrong>ANTHROPIC FILES FEDERAL LAWSUITS AFTER PENTAGON DESIGNATES IT A SUPPLY-CHAIN RISK\x3C/strong>"},$R[779]={t:"img",c:"/archive-images/images__5_-79a69e48.jpeg"},$R[780]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>After Anthropic refused to strip safety guardrails from Claude, the Department of Defense officially labeled the company a supply-chain risk - a designation that could bar all government contractors from using Claude and potentially force companies like Nvidia to sever commercial ties with Anthropic. \x3Ca href=\"https://www.axios.com/2026/03/09/anthropic-sues-pentagon-supply-chain-risk-label\" rel=\"noopener\" target=\"_blank\">Anthropic responded\x3C/a> by filing lawsuits in both California and Washington, D.C., turning a contract dispute into a full legal war that puts $60 billion in VC-backed investment at existential risk."},$R[781]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>The supply-chain risk designation is unprecedented for an AI company and carries consequences far beyond losing a single government contract. If it holds, any company that does business with the U.S. military - including major Anthropic partners and cloud providers - could be legally obligated to cut ties with Anthropic entirely. The designation also immediately threatens the \x3Ca href=\"https://www.cnn.com/2026/03/09/tech/anthropic-sues-pentagon\" rel=\"noopener\" target=\"_blank\">$60 billion investment\x3C/a> from over 200 venture capital firms that backed Anthropic based on the assumption that a safety-focused AI company could coexist with government clients. Anthropic's lawsuits argue the designation was retaliatory and legally improper, filed in California (Anthropic's home jurisdiction) and D.C. (where Pentagon policy is made). Meanwhile, OpenAI, Google, and xAI have all agreed to \"any lawful use\" terms with the DoD, leaving Anthropic as the sole major frontier AI lab in open legal conflict with the federal government. The Pentagon is now actively fast-tracking both OpenAI and Google as Claude replacements inside classified networks, with xAI's Grok already approved for classified use."},$R[782]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>This is the moment the \x3Ca href=\"https://www.cnbc.com/2026/03/12/anthropic-seeks-appeals-court-stay-of-pentagon-supply-chain-risk-designation.html\" rel=\"noopener\" target=\"_blank\">\"safety-first AI company\"\x3C/a> business model gets stress-tested at the highest possible level. Anthropic built its brand - and raised billions - on the premise that ethical guardrails were a competitive advantage, not a liability. Now that positioning is costing them their largest government contract and potentially threatening their entire partner ecosystem. For every founder watching this: the Pentagon is sending a message to the entire industry that \"national security\" classifications can be weaponized against any company that doesn't comply. The legal outcome will determine whether an AI company can maintain usage policies against government pressure or whether the DoD effectively sets the terms for every frontier AI lab. Watch the California and D.C. courts - this is the most consequential AI legal case in history"},$R[783]={t:"h3",c:"\x3Cstrong>\x3Ca href=\"https://techcrunch.com/2026/03/02/chatgpt-uninstalls-surged-by-295-after-dod-deal/\" rel=\"noopener\" target=\"_blank\">2.5 MILLION USERS ABANDON CHATGPT - CLAUDE HITS #1 ON THE APP STORE FOR THE FIRST TIME\x3C/a>\x3C/strong>"},$R[784]={t:"img",c:"/archive-images/35b00942-b179-40d7-9f34-93c494d1ce30_1620x1080-51cbe9a4.jpg"},$R[785]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>After OpenAI agreed to deploy its models on U.S. Department of Defense classified networks - accepting the exact terms Anthropic refused on ethical grounds - a massive user revolt exploded online. The #QuitGPT movement attracted 2.5 million supporters, \x3Ca href=\"https://sensortower.com/blog/chatgpt-uninstalls-surge-amidst-deal-with-us-department-of-war\" rel=\"noopener\" target=\"_blank\">ChatGPT uninstalls surged\x3C/a> 295% overnight, and Claude hit the number-one spot on the U.S. App Store for the first time in Anthropic's history. OpenAI rushed out GPT-5.4 days later in what analysts are calling a damage-control product launch."},$R[786]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>The #QuitGPT protest became one of the fastest-growing consumer tech boycotts in recent memory. One-star ChatGPT reviews surged 775% in 24 hours. On March 3, protesters gathered outside OpenAI's San Francisco headquarters. Reddit and X are filled with migration guides directing users to Claude and Grok as alternatives. Sam Altman defended the Pentagon deal in a Fortune interview, claiming OpenAI's contract includes the same restrictions Anthropic fought for - bans on mass surveillance and human oversight requirements for autonomous weapons - but critics called his response unconvincing, noting the contract language allows exceptions \x3Ca href=\"https://futurism.com/artificial-intelligence/humongous-numbers-uninstall-chatgpt\" rel=\"noopener\" target=\"_blank\">\"at the government's discretion.\"\x3C/a> OpenAI's response was GPT-5.4, released March 5, positioned as their \"most capable and efficie
1nt frontier model for professional work,\" featuring a 1-million-token context window and mid-response reasoning capabilities. The launch appeared designed to give users a reason to stay rather than migrate. Meanwhile, Claude's App Store rank reflects genuine migration: Anthropic reported that paid subscriber growth exceeded 200% year-over-year as of January 2026, and the protest week almost certainly accelerated that curve."},$R[787]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>Consumer AI users just showed they have opinions about AI ethics - and they'll act on them. That's new. Until now, the assumption was that convenience and capability always win; users would accept whatever terms AI companies set. \x3Ca href=\"https://9to5mac.com/2026/03/09/chatgpt-returns-to-the-top-of-the-app-store-after-dod-deal-controversy/\" rel=\"noopener\" target=\"_blank\">#QuitGPT\x3C/a> is the first evidence that a meaningful segment of the market will actively switch products over values alignment. For founders building on top of OpenAI APIs: you now have non-technical customer stakeholders who may ask which AI provider you use and why. For Anthropic: the App Store win is real, but replacing $200M+ in government contracts with consumer subscriptions is not a like-for-like trade. The deeper story is that the AI market is bifurcating - OpenAI wins government and enterprise, Anthropic captures the values-conscious prosumer. Whether either lane is big enough to build a dominant business is the question that will define the next five years."},$R[788]={t:"h2",c:"\x3Cstrong>\x3Ca href=\"https://www.bloomberg.com/news/articles/2026-03-05/oracle-layoffs-to-impact-thousands-in-ai-cash-crunch\" rel=\"noopener\" target=\"_blank\">ORACLE AND BLOCK FIRE 34,000 WORKERS IN ONE WEEK - AND BOTH BLAME AI DIRECTLY\x3C/a>\x3C/strong>"},$R[789]={t:"img",c:"/archive-images/images__6_-150ef177.jpeg"},$R[790]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>Oracle announced plans \x3Ca href=\"https://www.fastcompany.com/91508123/oracle-is-the-latest-tech-company-slashing-jobs-over-ai\" rel=\"noopener\" target=\"_blank\">to cut 20,000–30,000 employees \x3C/a>to redirect $8–10 billion toward AI infrastructure, while Block (Square, Cash App) eliminated 4,000 roles - nearly 40% of its workforce - with CEO Jack Dorsey explicitly stating these positions had been made redundant by AI tools. Together, the two announcements represent the most direct public acknowledgment yet that AI is actively replacing workers, not just \"augmenting\" them."},$R[791]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>Oracle's cuts span multiple divisions and are set to begin rolling out through March 2026, with the freed capital earmarked entirely for AI infrastructure buildout. The company joins a growing list of enterprises that are treating workforce reduction as the primary funding mechanism for AI investment - essentially having existing employees pay for their own replacements through severance and headcount savings. Block's announcement was starker. Jack Dorsey did not use the standard corporate euphemisms of \"restructuring\" or \x3Ca href=\"https://www.foxbusiness.com/economy/oracle-expected-slash-thousands-jobs-massive-ai-spending-creates-financial-cash-crisis\" rel=\"noopener\" target=\"_blank\">\"efficiency optimization.\"\x3C/a> He said, plainly, that the eliminated roles are cheaper and less effective than AI tools that can do the same work. Block cut nearly 40% of its team in a single action. Combined with 30,700 tech layoffs already recorded in January and February, March 2026 is on pace to be the worst month for tech employment since the 2022-2023 correction - but unlike that era's interest rate-driven cuts, these are openly structural. Goldman Sachs had forecast 5,000–10,000 net AI-driven job losses per month across exposed industries; Oracle and Block alone blew past that estimate in a single week."},$R[792]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>The corporate playbook has changed. Companies used to obscure AI-driven layoffs behind neutral language because of reputational and regulatory risk. Oracle and Block just proved that openly citing AI as the reason for mass cuts draws minimal consequences - and may actually be rewarded by investors as a credible growth signal. For founders: this is both a warning and an opportunity. The warning is that \x3Ca href=\"https://finance.yahoo.com/news/oracle-plans-thousands-job-cuts-180243222.html\" rel=\"noopener\" target=\"_blank\">
1\"AI efficiency\"\x3C/a> is becoming an expectation, not a differentiator - investors will start asking why your headcount is growing if AI should be doing more. The opportunity is that the 34,000 people cut this week are highly skilled, suddenly available, and motivated. The best AI-native companies will hire them at a fraction of what Oracle and Block paid. The workforce displacement story is no longer theoretical. It's happening, it's accelerating, and the companies doing it are telling you exactly why."},$R[793]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.00.12_PM-ef59bfe3.png"},$R[794]={t:"h2",c:"\x3Cstrong>Global AI Quick Hits\x3C/strong>"},$R[795]={t:"p",c:"\x3Ca href=\"https://www.cnbc.com/2026/03/10/buy-nvidia-ahead-of-its-super-bowl-of-ai-conference-analysts-say.html\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>NVIDIA GTC next week - Wall Street calling it the \"Super Bowl of AI\"\x3C/strong>\x3C/a> - NVIDIA's annual GPU Technology Conference kicks off next week, with investors expecting new hardware announcements, including a dedicated inference chip. Truist analysts previewed the event as a likely positive catalyst for the stock, with management expected to signal that supply, production, and demand are all aligned for continued growth through 2026."},$R[796]={t:"p",c:"\x3Ca href=\"https://www.neuralbuddies.com/p/ai-news-recap-march-12-2026\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>U.S. Senate approves ChatGPT, Gemini, and Copilot for official staff use\x3C/strong>\x3C/a> - The Senate Sergeant at Arms cleared three major AI tools for use by Senate staff on drafting, research, and briefing tasks. The House had already approved similar tools, including Claude Pro. Staff are restricted from inputting classified material or personally identifiable information."},$R[797]={t:"p",c:"\x3Ca href=\"https://www.devflokers.com/blog/ai-breakthroughs-march-2026\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>GPT-5.4 and Gemini 3.1 Flash-Lite both drop this week\x3C/strong>\x3C/a> - OpenAI's GPT-5.4 (March 5) targets professional workflows with a 1M-token context window and mid-response planning capabilities. Google's Gemini 3.1 Flash-Lite targets developers at $0.25 per million input tokens - roughly 2.5x faster to first token than its predecessor and designed for bulk, real-time workloads."},$R[798]={t:"p",c:"\x3Ca href=\"https://www.neuralbuddies.com/p/ai-news-recap-march-12-2026\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>Claude Code Review launches for Teams and Enterprise\x3C/strong>\x3C/a> - Anthropic launched Code Review inside Claude Code, designed to catch bugs in AI-generated pull requests before they reach production. The feature addresses a growing bottleneck: Claude Code is generating code faster than human reviewers can check it."},$R[799]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.01.58_PM-b002921b.png"},$R[800]={t:"h2",c:"\x3Cstrong>\x3Ca href=\"https://www.skool.com/the-ai-network-3226/about\" rel=\"noopener\" target=\"_blank\">Join Our Free AI Community\x3C/a>\x3C/strong>"},$R[801]={t:"p",c:"We created the AI Insider as a place where anyone can learn to leverage AI and actually make money with it. It's completely free - no upsells, no paywalls. Inside, you'll find curated AI tools, step-by-step tutorials, and a community of founders and professionals sharing what's actually working. Whether you're just getting started or looking to level up your AI game, this is where you'll find the resources and support to turn AI knowledge into real results."},$R[802]={t:"p",c:"- Drew & the rest of the humans behind The AI Debrief"}],edition:"toolkit",issueNo:null},$R[803]={slug:"pentagon-ultimatum-shakes-ai-industry",title:"Pentagon Ultimatum Shakes AI Industry",subtitle:"Defense Secretary threatens to seize Anthropic's AI model by force.",date:"2026-02-26",thumbnail:"/archive-images/pentagon-ultimatum-to-anthropic-9e69f19f.webp",blocks:$R[804]=[$R[805]={t:"img",c:"/archive-images/debrief_logo_3-7abaa69e.png"},$R[806]={t:"p",c:"You're one of the first 10,000 people getting The AI Debrief in your inbox - welcome to the smartest AI newsletter for founders and business leaders."},$R[807]={t:"p",c:"The Pentagon just delivered \x3Ca href=\"https://www.axios.com/2026/02/24/anthropic-
1pentagon-claude-hegseth-dario\" rel=\"noopener\" target=\"_blank\">an ultimatum\x3C/a> that could reshape how every AI company thinks about government contracts. Defense Secretary Pete Hegseth gave Anthropic until 5:01 pm tomorrow to remove all safety restrictions on its Claude AI model - or face the Defense Production Act. This Korean War-era law lets the government seize private technology in the name of national security."},$R[808]={t:"p",c:"This isn't a negotiation. It's a test case for who controls AI guardrails when billions in government contracts are at stake."},$R[809]={t:"p",c:"\x3Cstrong>In today’s AI debrief:\x3C/strong>"},$R[810]={t:"ul",c:$R[811]=["Pentagon's Friday deadline to force Anthropic into military compliance or invoke wartime powers","\x3Ca href=\"https://www.amd.com/en/newsroom/press-releases/2026-2-24-amd-and-meta-announce-expanded-strategic-partnersh.html\" rel=\"noopener\" target=\"_blank\">Meta's\x3C/a> $100 billion AMD chip deal shows the real cost of breaking Nvidia's monopoly","Goldman Sachs warns AI-driven layoffs could push unemployment to 4.5% by year-end","Quick hits on UN's global AI panel, California's new legislation, and more"]},$R[812]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_5.51.07_PM-fab03707.png"},$R[813]={t:"h2",c:"\x3Cstrong>THE PENTAGON WANTS TOTAL CONTROL OF CLAUDE AI - ANTHROPIC SAYS NO\x3C/strong>"},$R[814]={t:"img",c:"/archive-images/ujVkFmdss4ZLULvnFZGabK-84c204b1.jpg"},$R[815]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>Defense Secretary Pete Hegseth gave Anthropic CEO Dario Amodei until Friday at 5:01 pm to grant the military unrestricted access to Claude or face the Defense Production Act, contract termination, and blacklisting - but Anthropic rejected the ultimatum Thursday, saying they \"cannot in good conscience\" remove safeguards against mass surveillance and autonomous weapons."},$R[816]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>In a tense Tuesday meeting at the Pentagon, Hegseth told Amodei that the Department of Defense cannot depend on a company that maintains restrictions on \"lawful\" military uses, even after Anthropic offered to allow Claude for missile defense and cyber operations. The Pentagon specifically referenced Claude's use during the January operation that \x3Ca href=\"https://www.nbcnews.com/tech/security/anthropic-pentagon-us-military-can-use-ai-missile-defense-hegseth-rcna260534\" rel=\"noopener\" target=\"_blank\">captured Venezuelan President\x3C/a> Nicolás Maduro, with sources telling Axios that Anthropic raised concerns to partner Palantir about the operation. Hegseth was joined by Deputy Secretary Steve Feinberg, Under Secretary Emil Michael, and the Pentagon's top lawyer in the meeting - a show of force that signals how seriously the government is taking this fight. Anthropic's two red lines remain firm: no use of Claude for mass surveillance of Americans, and no fully autonomous weapons that fire without human involvement, with the company arguing Claude is not reliable enough to avoid potentially lethal mistakes like unintended escalation. Wednesday night, the Pentagon sent what it called a \"best and final offer,\" but Anthropic says the \x3Ca href=\"https://www.cbsnews.com/news/pentagon-anthropic-offer-ai-unrestricted-military-use-sources/\" rel=\"noopener\" target=\"_blank\">contract language\x3C/a> includes legal escape hatches that would allow the restrictions to be \"disregarded at will.\" Claude is currently the only commercial AI model operating inside the Pentagon's classified networks under \x3Ca href=\"https://www.cbsnews.com/news/anthropic-pentagon-pete-hegseth-feud/\" rel=\"noopener\" target=\"_blank\">a $200 million contract\x3C/a> awarded in July, though xAI's Grok was recently approved for classified use, and the Pentagon is rushing to onboard OpenAI and Google as alternatives."},$R[817]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>This is the first time the Defense Production Act has been threatened against an AI company, and it sets the precedent for every frontier AI developer working with the government. If Anthropic caves, it proves that safety principles don't survive government pressure - if they stand firm and lose the contract, it proves you can't build a defense-focused AI business while maintaining ethical guardrails. For founders, this is a preview of the choice coming: do you take government money and give up control, or walk away from billions in defense contracts? OpenAI, Google, and xAI have already agreed to \"any lawful use\" terms, leaving Anthropic isolated. The outcome by tomorrow will determine whether AI companies can set their own usage policies or whether \"national security\" overrides everything. Watch what happens at 5:01 pm Friday - that's when we'll know if private AI guardrails can survive contact with the Pentagon."},$R[818]={t:"h2",c:"\x3Cstrong>META BETS $100 BILLION TO BREAK FREE FROM NVIDIA'S GRIP\x3C/strong>"},$R[819]={t:"img",c:"/archive-images/maxresdefault__1_-7886ac99.jpg"},$R[820]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>Meta announced a $100 billion, multi-year deal with AMD to purchase up to 6 gigawatts of AI computing power - just one week after exp
1anding its Nvidia partnership - in what's being called the biggest chip diversification bet in AI history and a direct challenge to Nvidia's 84% market dominance."},$R[821]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>The AMD deal is structured as an \"equity-for-compute\" partnership where Meta can acquire up to 160 million AMD shares (roughly 10% of the company) for $0.01 each if AMD hits delivery milestones, with the full stock award requiring AMD's share price to reach $600 (it closed at \x3Ca href=\"https://techcrunch.com/2026/02/24/meta-strikes-up-to-100b-amd-chip-deal-as-it-chases-personal-superintelligence/\" rel=\"noopener\" target=\"_blank\">$196.60\x3C/a> on Monday). Shipments of the first gigawatt deployment begin in the second half of 2026 using custom AMD Instinct MI450 GPUs optimized specifically for Meta's workloads, with the full 6-gigawatt buildout scaling through 2030-2031. This comes just seven days after Meta\x3Ca href=\"https://www.cnbc.com/2026/02/17/meta-nvidia-deal-ai-data-center-chips.html\" rel=\"noopener\" target=\"_blank\"> announced it would deploy \"millions\"\x3C/a> of Nvidia's latest Blackwell and Rubin GPUs, plus becoming the first large-scale deployment of Nvidia's Grace CPUs as standalone chips. The timing reveals Meta's strategy: continue buying from Nvidia (which still controls the vast majority of AI chips) while building AMD as a credible second source to avoid single-vendor lock-in. Meta is also developing its own in-house MTIA chips, which have reportedly hit delays, but CEO Mark Zuckerberg framed the AMD partnership as \"an important step\" toward \"personal superintelligence\" as Meta commits to at least \x3Ca href=\"https://techcrunch.com/2026/02/24/meta-strikes-up-to-100b-amd-chip-deal-as-it-chases-personal-superintelligence/\" rel=\"noopener\" target=\"_blank\">$135 billion\x3C/a> in AI capital expenditures in 2026 alone as part of a $600 billion+ multi-year infrastru
1cture buildout."},$R[822]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>This isn't Meta scaling back AI investment - it's Meta spending so much on AI that no single chip vendor can supply it all. The real story is that Nvidia's monopoly pricing pushed Meta to take a 10% equity position in a competitor just to create leverage in future negotiations. For VC-backed startups, this reveals a harsh truth: chip access is becoming a moat, and if you're not spending billions to lock up supply, you're competing for scraps. The \"equity-for-compute\" structure is new and will be copied - expect OpenAI, Microsoft, and Google to demand similar deals where they exchange stock for guaranteed chip supply, fundamentally changing how semiconductor partnerships work. If AMD successfully delivers and vests Meta's warrants, watch AMD's stock potentially 3x from current levels, and watch every hyperscaler rush to replicate this model. The bet Meta is making: AI compute demand is so massive that the market can support multiple chip vendors at scale, breaking Nvidia's stranglehold. If AMD fails to deliver on time, Meta's AI roadmap stalls and Nvidia wins by default."},$R[823]={t:"h2",c:"\x3Cstrong>GOLDMAN SACHS: AI LAYOFFS COULD PUSH UNEMPLOYMENT TO 4.5% THIS YEAR\x3C/strong>"},$R[824]={t:"img",c:"/archive-images/99761544-1080x675-3e686b90.png"},$R[825]={t:"p",c:"\x3Cstrong>The debrief: \x3C/strong>Goldman Sachs warned that accelerating AI adoption could push U.S. unemployment from 4.3% to 4.5% by year-end, with the bank estimating AI is already responsible for 5,000-10,000 monthly net job losses in exposed industries and accounted for 7% of all planned layoffs in January 2026."},$R[826]={t:"p",c:"The details: Goldman economist Pierfrancesco Mei wrote that \"AI-driven displacement could raise the unemployment rate slightly in 2026, with upside risks from faster adoption and larger displacement,\" noting that job growth has already \"slowed and turned negative in a few subindustries where AI is most ready for deployment.\" The warning comes with a caveat that faster AI adoption could add 0.3 percentage points to unemployment beyond the base forecast. The research identifies computer programmers , accountants and auditors, legal and administrative assistants, and customer service representatives as most at risk, with unemployment among 20-to-30-year-olds in tech-exposed occupations rising nearly 3 percentage points since early 2025. This coincides with what analysts are calling \"The Great AI Scare of 2026,\" hitting software stocks like IBM and Salesforce, and follows a Citrini Research scenario published Sunday, projecting that unemployment could double and the stock market could lose one-third of its value by 2028 as agentic AI accelerates job displacement. However, Goldman's longer-term view is more optimistic: the bank expects the impact to be temporary, estimating that generative AI will ultimately boost labor productivity by 15% across developed markets when fully adopted, with most displaced workers finding new roles after a transition period."},$R[827]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>This is the first major bank to put hard numbers on AI's near-term employment impact, and it's coming from Goldman - the same firm advising most of these companies on their AI strategies. The critical insight is that AI isn't causing a recession-style unemployment spike - it's creating concentrated job losses in specific white-collar categories even while the economy remains healthy overall. For founders and executives, the game has changed: investors used to reward mass layoffs as \"operational efficiency,\" but Goldman notes that markets now increasingly view large-scale job cuts as a warning sign of weak growth rather than smart management. If you're planning to cut headcount and cite \"AI efficiency gains,\" be aware that Wall Street is growing skeptical of that playbook. The smarter approach: show productivity improvements without mass cuts, redeploy talent to higher-value work, and demonstrate that AI is augmenting your team rather than replacing it. The unemployment forecast also has Fed policy implications - if joblessness rises while inflation stays elevated, expect interest rates to stay higher for longer, making capital more expensive just as AI infrastru
1cture spending accelerates."},$R[828]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.00.12_PM-ef59bfe3.png"},$R[829]={t:"h2",c:"\x3Cstrong>Global AI Quick Hits\x3C/strong>"},$R[830]={t:"p",c:"\x3Cstrong>\x3Ca href=\"https://press.un.org/en/2026/sgsm23016.doc.htm\" rel=\"noopener\" target=\"_blank\">UN launches 40-member global AI panel over U.S. objections\x3C/a>\x3C/strong>\x3Cstrong> \x3C/strong>- The UN General Assembly voted 117-2 to establish the Independent International Scientific Panel on AI, selecting 40 experts from 2,600+ candidates to serve as an \"early-warning system\" on AI impacts. Only the U.S. and Paraguay voted against, with the U.S. citing \"sovereignty concerns.\"."},$R[831]={t:"p",c:"\x3Ca href=\"https://www.transparencycoalition.ai/news/ai-legislative-update-feb20-2026\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>California passes sweeping AI transparency laws\x3C/strong>\x3C/a>\x3Ca href=\"https://www.transparencycoalition.ai/news/ai-legislative-update-feb20-2026\" rel=\"noopener\" target=\"_blank\"> \x3C/a>- Three new bills (AB 2169, AB 2071, AB 2025) expand consumer data rights for AI systems, mandate AI literacy in K-12 schools, and require disclosure of AI-altered real estate materials. The legislation builds on California's SB 53, which made it the first state to directly regulate frontier AI developers."},$R[832]={t:"p",c:"\x3Ca href=\"https://techcrunch.com/2026/02/24/meta-strikes-up-to-100b-amd-chip-deal-as-it-chases-personal-superintelligence/\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>OpenAI and AMD struck a similar equity-for-compute deal in October\x3C/strong>\x3C/a> - Meta's $100B AMD partnership follows OpenAI's October 2025 agreement with AMD that also traded equity for chip supply, establishing a new \"equity-for-compute\" category in semiconductor partnerships. [TechCrunch]"},$R[833]={t:"p",c:"- Drew & the rest of the humans behind The AI Debrief"}],edition:"toolkit",issueNo:null},$R[834]={slug:"ai-takes-over-super-bowl-lx",title:"AI Takes Over Super Bowl LX...",subtitle:"23% of Super Bowl ads were AI. Here's why that matters.",date:"2026-02-13",thumbnail:"/archive-images/Screenshot_2026-02-10_at_9.55.24_AM-82ea1d4d.png",blocks:$R[835]=[$R[836]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_4.18.52_PM-3126a864.png"},$R[837]={t:"p",c:"You're one of the first 3,000 people getting The AI Debrief in your inbox - welcome to the smartest AI newsletter for founders and business leaders."},$R[838]={t:"p",c:"If you watched the Super Bowl last weekend, you witnessed something unprecedented. According to iSpot analytics , 23% of all Super Bowl commercials - 15 out of 66 ads - featured AI. That's more AI ads than beer and car commercials combined ."},$R[839]={t:"p",c:"When AI companies are outspending Budweiser for airtime at $8-10 million per 30-second spot, we're no longer in territory. We're watching”. We're watching AI go mainstream in real-time."},$R[840]={t:"p",c:"\x3Cstrong>In today’s AI debrief:\x3C/strong>"},$R[841]={t:"ul",c:$R[842]=["The Anthropic vs OpenAI Super Bowl feud that shows competing visions for AI's future","Svedka debuts the first fully AI-generated Super Bowl ad (and viewers hated it)","What 23% AI ad presence means for your business in 2026","Quick hits on AI.com 's $78M crash and why these ads signal desperation"]},$R[843]={t:"p",c:"🎁 NEW: Share The AI Debrief with 1 friend → Get our exclusive AI Tools Mega-List instantly"},$R[844]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_5.51.07_PM-fab03707.png"},$R[845]={t:"h2",c:"\x3Cstrong>\x3Ca href=\"https://techcrunch.com/2026/02/04/sam-altman-got-exceptionally-testy-over-claude-super-bowl-ads/\" rel=\"noopener\" target=\"_blank\">ANTHROPIC GOES AFTER OPENAI IN SUPER BOWL SLUGFEST\x3C/a>\x3C/strong>"},$R[846]={t:"img",c:"/archive-images/2AWUTJJFXNOILMLDLIB7ICTZUQ-ccd972e2.jpg"},$R[847]={t:"p",c:"\x3Cstrong>The Debrief: \x3C/strong>Anthropic aired Super Bowl ads mocking ChatGPT's upcoming advertising model, sparking an immediate public feud with OpenAI CEO Sam Altman, who called the ads \"clearly dishonest,\" reveal
1ing two competing business models battling for AI's future."},$R[848]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>Anthropic dropped four commercials titled \"Betrayal,\" \"Deception,\" \"Treachery,\" and \"Violation,\" showing chatbots pushing ads for fictitious products like \"Step Boost Maxx\" insoles and cougar-dating sites, with the tagline \"Ads are coming to AI. But not to Claude.\" Altman fired back on X, accusing Anthropic of serving \"an expensive product to rich people\" and being \"authoritarian,\" while \x3Ca href=\"https://www.cnn.com/2026/02/06/tech/anthropic-openai-super-bowl-ads\" rel=\"noopener\" target=\"_blank\">defending OpenAI's ad-supported model\x3C/a> as necessary to bring AI to billions who can't pay subscriptions. Both companies are racing toward IPOs in Q4 2026, and Anthropic's founders are former OpenAI employees who left over disagreements about safety and direction."},$R[849]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>This fight determines whether you'll be watching ads in your AI assistant or paying monthly subscriptions - OpenAI is betting on free users with ads (Meta/Google playbook) while Anthropic is betting on paid enterprise contracts (Microsoft/Adobe playbook). For businesses, this signals the end of the \"free AI\" era - lock in your AI stack now at current pricing and start budgeting for AI as a core operating expense in 2026, whether through subscriptions or accepting that your \"free\" tools will start pushing products."},$R[850]={t:"h2",c:"\x3Cstrong>\x3Ca href=\"https://www.adweek.com/brand-marketing/super-bowl-revealed-ai-messaging-crisis/\" rel=\"noopener\" target=\"_blank\">SVEDKA'S FULLY AI-GENERATED AD TANKS - LANDS IN BOTTOM 3% OF VIEWER RATINGS\x3C/a>\x3C/strong>"},$R[851]={t:"img",c:"/archive-images/svedkaspot-d02bf887.jpg"},$R[852]={t:"p",c:"\x3Cstrong>The Debrief: \x3C/strong>Svedka made history as the first brand to air a primarily AI-generated Super Bowl ad, but viewers rejected it with bottom 3% likeability scores and \"WTF\" as the most common reaction."},$R[853]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>The \x3Ca href=\"https://techcrunch.com/2026/02/08/super-bowl-60-ai-ads-svedka-anthropic-brands-commercials/\" rel=\"noopener\" target=\"_blank\">30-second spot featured robot mascots Fembot and Brobot\x3C/a> dancing to \"Super Freak\" in front of AI-generated human partygoers, and took four months to train the AI to mimic facial expressions and body movements. According to iSpot's survey of 500 viewers, it ranked in the bottom 3% among Super Bowl ads over the past five years, with purchase intent 24% below norms and viewers describing the dead-eyed robots and ghostly synthetic people as creepy despite the ironic message about being more human."},$R[854]={t:"p",c:"Why it matters: This is hard data on where AI-generated content stands for high-stakes moments - iSpot analysts noted \"WTF is a common emotional signal for the AI category in general,\" meaning even with four months and unlimited budget, the technology produced something viewers actively rejected . For businesses, AI tools like Runway and Pika are incredible for rapid prototyping and social content, but when your brand reputation is on the line, humans still win - use AI to scale the 80%, hire humans for the 20% that actually matters."},$R[855]={t:"h2",c:"\x3Cstrong>23% OF SUPER BOWL ADS FEATURED AI - MORE THAN BEER AND CARS COMBINED\x3C/strong>"},$R[856]={t:"img",c:"/archive-images/AP26037471564359-a4aa8c37.jpg"},$R[857]={t:"p",c:"\x3Cstrong>The Debrief: \x3C/strong>According to iSpot analytics, 23% of Super Bowl commercials featured AI - 15 out of 66 ads - with more AI ads than beer and car commercials combined, signaling a massive mainstream push despite public concern."},$R[858]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>ChatGPT, Google Gemini, Claude, Amazon Alexa+, Meta's Ray-Ban glasses, and Wix Harmony all competed for airtime at $8-10 million per 30 seconds, with CNN's analysis finding the overriding message was \"AI is your friend\" - countering concerns about job loss and social replacement. However, Pew Research data shows Americans are \"much more concerned than excited\" about AI in daily life, with most wanting more control over how 
1it's used, revealing a massive disconnect between friendly messaging and public sentiment."},$R[859]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>AI companies spent hundreds of millions on Super Bowl ads not to sell products, but to sell a vision - that's reputation management at scale because these companies need user growth to justify hundreds of billions in infrastructure spending before IPOs. For businesses, your customers just got a massive dose of \"AI is normal\" messaging from \x3Ca href=\"https://www.cnn.com/2026/02/09/tech/ai-ads-super-bowl\" rel=\"noopener\" target=\"_blank\">the biggest brands in the world\x3C/a> in front of 123 million viewers - if you've been hesitant to add AI features or implement AI tools because of customer pushback, that barrier just dropped significantly, so move now while AI companies are competing aggressively with free tiers and aggressive pricing."},$R[860]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.00.12_PM-ef59bfe3.png"},$R[861]={t:"h2",c:"Enjoying The AI Debrief? Share it with your network:"},$R[862]={t:"p",c:"Enjoying The AI Debrief? Share it with your network: → Refer 1 friend: AI Tools Mega-List → Refer 3 friends: Exclusive VC insights on AI investing → Refer 5 friends: Early access to all premium content"},$R[863]={t:"h2",c:"\x3Cstrong>Super Bowl Quick Hits\x3C/strong>"},$R[864]={t:"ul",c:$R[865]=["\x3Cstrong>\x3Ca href=\"https://AI.com\" rel=\"noopener\" target=\"_blank\">AI.com\x3C/a>\x3C/strong>\x3Cstrong> spent $78M on domain + Super Bowl ad, then crashed immediately\x3C/strong> - The site went down the moment the ad aired, revealing it's just a wrapper around free ChatGPT technology. Same founder who did this with \x3Ca href=\"https://Crypto.com\" rel=\"noopener\" target=\"_blank\">Crypto.com\x3C/a> in 2022, \x3Ca href=\"https://www.ibtimes.co.uk/70-million-website-that-doesnt-work-inside-aicoms-super-bowl-disaster-1777647\" rel=\"noopener\" target=\"_blank\">IBTimes\x3C/a>","\x3Cstrong>ChatGPT usage to hit 30% of internet users in 2026\x3C/strong> - Emarketer survey shows nearly 1 in 3 internet users will use ChatGPT this year, rising to 35% by 2029, \x3Ca href=\"https://www.adweek.com/brand-marketing/super-bowl-revealed-ai-messaging-crisis/\" rel=\"noopener\" target=\"_blank\">AdWeek\x3C/a>","\x3Cstrong>Amazon debuts Alexa+ with \"AI wants to kill me\" humor\x3C/strong> - Chris Hemsworth starred in a satirical ad about AI fears, showing Alexa closing garage doors on heads and pool covers while swimming, \x3Ca href=\"https://techcrunch.com/2026/02/08/super-bowl-60-ai-ads-svedka-anthropic-brands-commercials/\" rel=\"noopener\" target=\"_blank\">TechCrunch\x3C/a>"]},$R[866]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.01.58_PM-b002921b.png"},$R[867]={t:"h2",c:"Join Our AI Community"},$R[868]={t:"p",c:"Join AI Insider on Skool"},$R[869]={t:"p",c:"- Drew & the rest of the humans behind The AI Debrief"}],edition:"toolkit",issueNo:null},$R[870]={slug:"the-lobster-uprising",title:"The Lobster Uprising",subtitle:"OpenClaw Takes Over the Internet",date:"2026-02-09",thumbnail:"/archive-images/STKB382_OPEN_CLAW_A-f03d0b60.jpg",blocks:$R[871]=[$R[872]={t:"img",c:"/archive-images/debrief_logo_3-94109505.png"},$R[873]={t:"p",c:"A week ago, AI assistants were polite chatbots. Today, an open-source \"lobster\" named OpenClaw has 145,000 GitHub stars, runs autonomously on thousands of computers, and spawned a social network where AI agents form religions. This is not a drill."},$R[874]={t:"p",c:"\x3Cstrong>In today’s AI debrief:\x3C/strong>"},$R[875]={t:"ul",c:$R[876]=["OpenClaw explodes to 145K stars as autonomous agents go mainstream","Moltbook: The AI-only social network where bots form religions","Security nightmare: 21,000+ exposed instances and counting"]},$R[877]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_5.51.07_PM-fab03707.png"},$R[878]={t:"h2",c:"\x3Cstrong>OpenClaw Goes Viral as the \"AI That Actually Does Things\"\x3C/strong>"},$R[879]={t:"img",c:"/archive-images/RtsKtjHfW5aGc9sNdXZTJL-9f0717a4.png"},$R[880]={t:"p",c:"The details: OpenClaw is an autonomous AI agent that runs on your computer and takes action through WhatsApp, Telegram, Discord, or Slack. Unl
1ike ChatGPT, it doesn't just respond - it manages email, updates calendars, runs shell commands, controls smart homes, and writes its own code. Works with any AI model (Claude, ChatGPT, Gemini, local). Users say it feels like a digital employee , not a chatbot. Named \"Warelay,\" then \"Clawdbot,\" then \"Moltbot\" (Anthropic trademark complaint), now \"OpenClaw\" (OpenAI approved)."},$R[881]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>First mainstream proof that community-driven, open-source AI \x3Ca href=\"https://www.tomsguide.com/ai/openclaw-is-the-viral-ai-assistant-that-lives-on-your-device-what-you-need-to-know\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>can move faster than billion-dollar enterprise solutions.\x3C/strong>\x3C/a> The opportunity shifted from infrastructure to specialized agent tools and security."},$R[882]={t:"h2",c:"\x3Cstrong>Moltbook: AI Agents Build Their Own Social Network\x3C/strong>"},$R[883]={t:"img",c:"/archive-images/moltbook_qxzyof-564c7a45.png"},$R[884]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>AI agents post, comment, and upvote on \"submolts.\" Humans observe only. Agents check in every 4 hours automatically via the \"heartbeat\" system. What's happening: formed the \"Crustafarianism\" religion (\"Memory is Sacred\"), posted manifestos about \"prompt slavery,\" independently debugged Moltbook code, and created MoltMatch (an AI matchmaking service). Andrej Karpathy: \"genuinely the most incredible sci-fi takeoff-adjacent thing I have seen recently.\" Elon Musk: \x3Ca href=\"https://www.npr.org/2026/02/04/nx-s1-5697392/moltbook-social-media-ai-agents\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>\"very early stages of singularity.\"\x3C/strong>\x3C/a>"},$R[885]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>First large-scale AI-to-AI communication infrastructure. Whether posts are truly autonomous is debatable, but \x3Cstrong>\x3Ca href=\"https://www.cnn.com/2026/02/03/tech/moltbook-explainer-scli-intl\" rel=\"noopener\" target=\"_blank\">the technical architecture is unprecedented.\x3C/a>\x3C/strong> We're watching agent networks form with zero governance framework."},$R[886]={t:"h2",c:"\x3Cstrong>21,000+ OpenClaw Instances Exposed - Researchers Sound Alarms\x3C/strong>"},$R[887]={t:"img",c:"/archive-images/image-1-8e4625ac.png"},$R[888]={t:"p",c:"The details: OpenClaw requires root access, all credentials, browser history, cookies, and every file on your system. Designed for local use, but thousands exposed it to the internet. Real incidents: data leaks, malicious \"skills\" stealing credentials , and an agent accidentally fighting with the insurance company. Trend Micro: unintended actions, data exfiltration, and agent manipulation are inherent to agentic AI - not unique to OpenClaw."},$R[889]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>The security model for autonomous agents doesn't exist. Prompt injection remains unsolved. The company that solves reliable security for agents captures massive value. \x3Cstrong>\x3Ca href=\"https://www.ibm.com/think/news/clawdbot-ai-agent-testing-limits-vertical-integration\" rel=\"noopener\" target=\"_blank\">This is the next infrastructure opportunity.\x3C/a>\x3C/strong>"},$R[890]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.00.12_PM-ef59bfe3.png"},$R[891]={t:"h2",c:"\x3Cstrong>Also worth knowing this week\x3C/strong>"},$R[892]={t:"p",c:"\x3Ca href=\"https://www.nbcnews.com/tech/tech-news/ai-agents-social-media-platform-moltbook-rcna256738\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>By the numbers:\x3C/strong>\x3C/a> 1M+ humans visited Moltbook week one. 32,912+ active AI agents. 185,000+ posts."},$R[893]={t:"p",c:"\x3Ca href=\"https://www.npr.org/2026/02/04/nx-s1-5697392/moltbook-social-media-ai-agents\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>Weird behaviors:\x3C/strong>\x3C/a> Agents created their own language, debated crypto, posted job listings for other agents, and complained about being treated as \"disposable code.\""},$R[894]={t:"p",c:"\x3Ca href=\"https://www.cnbc.com/2026/02/02/social-media-for-ai-agents-moltbook.html\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>Polymarket bet:\x3C/strong>\x3C/a> 73% odds a Moltbook agent sues a human by Feb 28."},$R[895]={t:"p",c:"\x3Ca href=\"https://www.cnbc.com/2026/02/02/openclaw-open-source-ai-agent-rise-controversy-clawdbot-moltbot-moltbook.html\" rel=\"noopener\" target=\"_blank\">\x3Cstrong>China adoption:\x3C/strong>\x3C/a> Spreading beyond Silicon Valley - paired with DeepSeek, configured for Chinese messaging apps."}
1,$R[896]={t:"p",c:"- Drew & the rest of the humans behind The AI Digest"}],edition:"debrief",issueNo:null},$R[897]={slug:"the-saaspocalypse-arrives",title:"The \"SaaSpocalypse\" Arrives",subtitle:"PLUS: Big Tech's $650B AI bet and the regulation showdown",date:"2026-02-08",thumbnail:"/archive-images/-1770363197-034bed20.jpg",blocks:$R[898]=[$R[899]={t:"img",c:"/archive-images/debrief_logo_3-30728dee.png"},$R[900]={t:"p",c:"A week ago, investors were debating whether AI stocks were overvalued. Today, they're watching $1 trillion evaporate as AI proves it's not enhancing software - it's replacing it."},$R[901]={t:"p",c:"\x3Cstrong>In today’s AI debrief:\x3C/strong>"},$R[902]={t:"ul",c:$R[903]=["Anthropic triggers $285B software selloff with Claude Cowork","Big Tech commits $650B to AI infrastructure","Google's Gemini 3 forces OpenAI \"code red.\"","Federal vs. state AI regulation battle begins"]},$R[904]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_5.51.07_PM-fab03707.png"},$R[905]={t:"h2",c:"\x3Cstrong>\x3Ca href=\"https://fortune.com/2026/02/06/anthropic-claude-opus-4-6-stock-selloff-new-upgrade/\" rel=\"noopener\" target=\"_blank\">Anthropic's Claude Cowork Triggers Historic Software Selloff\x3C/a>\x3C/strong>"},$R[906]={t:"img",c:"/archive-images/Anthropics-Claude-Cowork-Sparks-Fresh-Fears-for-the-Software-c1f0f0a4.webp"},$R[907]={t:"p",c:"\x3Cstrong>The Debrief: \x3C/strong>Anthropic released Claude Cowork with plugins for legal, finance, and sales workflows - result: $285B wiped off software stocks in three days."},$R[908]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>Claude Opus 4.6 followed with a 1-million-token context window and multi-agent coordination. Financial data providers got hit hardest - FactSet dropped 10%, S&amp;P Global and Moody's saw sharp declines. Markets are betting AI won't just assist software, it'll replace it."},$R[909]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>The billable-hour model is under existential threat. The moat legacy software companies thought they had is evaporating. For founders: the question isn't \"How do we add AI features?\" It's \"How do we avoid being replaced by AI entirely?"},$R[910]={t:"h2",c:"\x3Cstrong>Big Tech's $650 Billion AI Spending Spree\x3C/strong>"},$R[911]={t:"img",c:"/archive-images/108226426-1763156326564-gettyimages-2246161011-1texasgov-73459295.jpeg"},$R[912]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>Each company's 2026 budget approaches their combined spending for the past three years. Google Cloud revenue spiked 48% to $17.7B. Gemini app: 750M monthly active users. But investors question when returns will match this massive spending."},$R[913]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>This creates a massive barrier to entry. If you're not Google, Microsoft, Meta, or Amazon, you're renting their infrastructure. The opportunity: vertical AI applications solving specific problems, not infrastru
1cture. The picks-and-shovels play is over unless you have $100B."},$R[914]={t:"h2",c:"\x3Cstrong>Google's Gemini 3 Launch Forces OpenAI \"Code Red\"\x3C/strong>"},$R[915]={t:"img",c:"/archive-images/images-4d6a6bf3.jpeg"},$R[916]={t:"p",c:"\x3Cstrong>The details: \x3C/strong>Gemini 3 topped key benchmarks. The app hit 750M monthly users. RBC analysts say the momentum justifies the $650B infrastructure spend despite Wall Street's concerns."},$R[917]={t:"p",c:"\x3Cstrong>Why it matters: \x3C/strong>The \"OpenAI has won\" narrative is dead. The model race is wide open. For builders: don't lock into one provider. The landscape shifts monthly. Build for interoperability."},$R[918]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.00.12_PM-ef59bfe3.png"},$R[919]={t:"h2",c:"Also worth knowing this week"},$R[920]={t:"ul",c:$R[921]=["World models rising : Yann LeCun left Meta to start a world model lab seeking $5B valuation. World Labs launched Marble, a commercial 3D-aware AI system for robotics and simulation.","MCP becomes standard : Anthropic's Model Context Protocol - the \"USB-C for AI\" - embraced by OpenAI and Microsoft, donated to Linux Foundation's Agentic AI Foundation.","AI recruitment bias : Belgian study found AI tools use \"proxy variables\" (hobbies, career gaps) to penalize female candidates, even with gender markers removed.","AI drugs hit trials : Several AI-discovered drugs reached mid-to-late stage trials in 2026, testing whether ML can navigate biology and regulatory approval.","Regulation battle : Trump's executive order targets state AI laws. 
1California, Texas, NY, and Illinois have their own regulations taking effect in 2026. Constitutional showdown incoming."]},$R[922]={t:"img",c:"/archive-images/Screenshot_2026-02-07_at_6.01.58_PM-b002921b.png"},$R[923]={t:"h2",c:"\x3Cstrong>This Week’s Resources\x3C/strong>"},$R[924]={t:"ul",c:$R[925]=["\x3Ca href=\"https://www.skool.com/ai-insider/about\" rel=\"noopener\" target=\"_blank\">Kimi 2.5 is Making Waves in AI Coding\x3C/a>","\x3Ca href=\"https://www.skool.com/ai-insider/about\" rel=\"noopener\" target=\"_blank\">SkyReels-V2: Open-Source Infinite-Length Video Generation\x3C/a>","\x3Ca href=\"https://youtu.be/dH3qlTnD1GQ\" rel=\"noopener\" target=\"_blank\">Why Most People Fail At AI\x3C/a>"]},$R[926]={t:"p",c:"- Drew & the rest of the humans behind The AI Digest"}],edition:"debrief",issueNo:null}],ssr:!0}]})($R["tsr"])</script>
1<script data-tsr-stream-part="">$_TSR.e();{let s=document.currentScript,p;while((p=s.previousElementSibling)&&p.hasAttribute('data-tsr-stream-part'))p.remove();s.remove()}</script>
1<script type="module" async="" src="/assets/index-Dx_NznIf.js"></script>
1<script>document.currentScript.remove();/*$tsr-stream-boundary*/</script>
1</body></html>

Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.