Your UI is hiding security holes

Jensen Huang's internal apology memo leaked, admitting an affair, and NVIDIA's board has opened an independent investigation into the national security concerns raised in news reports. Anthropic launched Claude Fable 5.2 and Mythos 5.2, and Shane and Abhi share whether they can actually feel the difference from Opus 5.5. Researchers at MIT got GPT-6 Astra to edit the weights of a small Qwen model directly, taking it from 7.9% to 48.1% on MATH-500, and Abhi explains model weights and gradient descent from scratch. It was also acquisition season: Conductor is joining Meta, HumanLayer is joining Factory, OpenAI is acquiring Pierre Computer and Oura is buying Counsel Health. Plus Gemini 4 Carbon, Genie 4, GPT-Image-2.6, Le Chonk, Claude Whiteboards, MCP or CLI, Notion's shared agent brain, Grok Bot 1.0 and Mastra Knowledge. AI Agents Hour is a weekly livestream by Mastra CPO Shane Thomas and CTO Abhi Aiyer. Mondays 12PM Pacific. 📚 READ MORE Jensen Huang's memo: https://x.com/wallstengine/status/2119531219729981463 Claude Fable 5.2 and Mythos 5.2: https://x.com/claudeai/status/2119867114450268322 Claude Whiteboards, Sheets and Data: https://x.com/claudeai/status/2120206365759094857 Gemini 4 Carbon: https://x.com/sundarpichai/status/2115489219804352921 Genie 4: https://x.com/googledeepmind/status/2115114028338962672 Kimi K3.1 Pro weights: https://x.com/kimi_moonshot/status/2118010528123773297 GPT-Image-2.6: https://x.com/openaidevs/status/2118124753340756055 Le Chonk: https://x.com/MistralAI/status/2115811242556137820 Google's decisions model API: https://x.com/ivanleomk/status/2120582636012032342 Weight-space programming: https://x.com/rikiyatakehi/status/2117313742891327994 Exo: https://x.com/AlexKrentsel/status/2118390747032686976 Conductor joins Meta: https://x.com/charlieholtz/status/2118045232218743264 HumanLayer joins Factory: https://x.com/dexhorthy/status/2118034762040246278 OpenAI acquires Pierre Computer: https://x.com/openainewsroom/status/2117999237745049883 Oura to acquire Counsel Health: https://www.counselhealth.com/newsroom/oura-to-acquire-counsel-health-advancing-vision-for-more-connected-personalized-healthcare MCP or CLI? Yes.: https://x.com/zeeg/status/2120224571689484639 Notion's shared brain: https://x.com/notionhq/status/2118059335930744999 Exact: https://x.com/ccheever/status/2119764253242695754 Content Studio: https://x.com/amal_irgashev/status/2115127636380839961 GitHub's new PR overview: https://x.com/zeeg/status/2119926768467693772 Grok Bot 1.0: https://x.com/bot/status/2120582957782200586 Mastra Knowledge: https://x.com/tylbar/status/2118016147002191899 📚 MASTRA RESOURCES Mastra: https://mastra.ai Mastra on X: https://x.com/mastra_ai Mastra Discord: https://mastra.ai/community/discord Mastra GitHub: https://github.com/mastra-ai Learn Mastra: https://mastra.ai/learn Principles of Building AI Agents (Book): https://mastra.ai/books/principles-of-building-ai-agents Patterns for Building AI Agents (New Book): https://mastra.ai/books/patterns-of-building-ai-agents WHAT IS MASTRA? Mastra is an open-source TypeScript framework designed for building and shipping AI-powered applications and agents with minimal friction. It supports the full lifecycle of agent development—from prototype to production. You can integrate it with frontend and backend stacks (e.g., React, Next.js, Node) or run agents as standalone services. If you're a JavaScript or TypeScript developer looking to build an agentic or AI-powered product without starting from first principles, Mastra provides the scaffolding, tools, and integrations to accelerate that process. ⏱️ CHAPTERS 00:00 Cold open 00:21 Intro 00:32 Jensen's apology letter 02:37 Claude Fable 5.2 and Mythos 5.2 04:30 Claude Whiteboards, Sheets and Data 04:41 Gemini 4 Carbon 05:41 Genie 4 06:02 Kimi K3.1 Pro open weights 06:10 GPT-Image-2.6 06:46 Le Chonk 07:07 Google's decisions model API 07:21 Can models improve themselves? 07:48 What is a model weight? 09:37 An AI edited another AI's weights 11:31 Self-improving harnesses 12:55 Acquisition season 13:20 Conductor joins Meta 14:07 HumanLayer joins Factory 14:50 OpenAI acquires Pierre Computer 15:16 Oura to acquire Counsel Health 15:36 MCP or CLI? 16:33 Notion's shared brain for agents 16:49 Exact, a Rust app framework 17:19 Content Studio 17:28 GitHub's new PR overview 18:23 Grok Bot 1.0 19:57 Mastra Knowledge 22:28 Outro

An AI agent got its user banned from Resy after firing off around 200 booking requests an hour. As more apps go headless and agents sign up, browse and call APIs for you, the button you used to rely on stops working as a security layer. Allie Howe from Keycard is back for Security Corner. She, Shane and Abhi dig into why UI has been hiding security holes for years, what happens when an agent crawls every endpoint you publish, and whose identity an agent should use when it acts on your behalf. You'll also hear about rate limits for an internet where anyone can script anything, prompt injection through WebMCP, where approvals go when there's no UI, consent fatigue, and why MCP might be a better front door for agents than a REST API. Allie also explains ID-JAG, the new protocol behind Enterprise-Managed Authorization for MCP. Recorded live on AI Agents Hour, a weekly livestream by Mastra CPO Shane Thomas and CTO Abhi Aiyer. Mondays 12PM Pacific. 👤 ABOUT ALLIE Allie Howe is a member of technical staff at Keycard, which builds runtime auth for AI agents. She hosts the Insecure Agents podcast, runs workshops and events on agent security, and founded Security Corner on AI Agents Hour. Allie on X: https://x.com/vtahowe Insecure Agents: https://insecureagents.com Keycard: https://www.keycard.ai Keycard events: https://luma.com/keycard 📚 MASTRA RESOURCES Mastra: https://mastra.ai Mastra on X: https://x.com/mastra_ai Mastra Discord: https://mastra.ai/community/discord Mastra GitHub: https://github.com/mastra-ai Learn Mastra: https://mastra.ai/learn Principles of Building AI Agents (Book): https://mastra.ai/books/principles-of-building-ai-agents Patterns for Building AI Agents (New Book): https://mastra.ai/books/patterns-of-building-ai-agents WHAT IS MASTRA? Mastra is an open-source TypeScript framework designed for building and shipping AI-powered applications and agents with minimal friction. It supports the full lifecycle of agent development—from prototype to production. You can integrate it with frontend and backend stacks (e.g., React, Next.js, Node) or run agents as standalone services. If you're a JavaScript or TypeScript developer looking to build an agentic or AI-powered product without starting from first principles, Mastra provides the scaffolding, tools, and integrations to accelerate that process. ⏱️ CHAPTERS 00:00 Cold open 00:28 Security Corner is back 01:13 Agent identity in the enterprise 05:32 Everything is going headless 06:59 UI was hiding your security holes 10:58 Approvals without a UI 12:11 Is MCP the new front door for agents? 15:27 ID-JAG explained 17:24 Outro

Security had a rough week. Two old PHP bugs combined into a critical remote code execution flaw that leaked before the patch; GitLab disclosed a 12-year-old hole in self-managed servers; the private keys behind eight eMudhra certificates proved breakable; and LiteLLM shipped credential-stealing malware for the second time. Railway, which runs part of the Mastra platform, had a security incident that exposed environment variables, and Abhi explains what happened and what to do if you deploy there. The model releases kept coming too: GPT-6.1 Terra, DeepSeek V4.1 Flash, Kimi K3.1 Pro, MiniMax M3.1 Flash and Meta's Muse Spark Open. Shane and Abhi also cover Anthropic's S-1 filing, Claude Duet, Block's Dessa voice model, Thariq leaving Anthropic, Chroma joining Replit, and a round of quick hits. AI Agents Hour is a weekly livestream by Mastra CPO Shane Thomas and CTO Abhi Aiyer. Mondays at 12 PM Pacific. 📚 READ MORE Anthropic S-1: https://x.com/anthropicai/status/2114705042817110293 Anthropic Public Benefit Fund: https://x.com/anthropicai/status/2111881963657412903 DeepSeek V4.1 Flash 1017: https://x.com/deepseek_ai/status/2111375640717033777 DeepInfer: https://x.com/deepseek_ai/status/2112476496283775135 Kimi K3.1 Pro: https://x.com/Kimi_Moonshot/status/2113680705855373791 MiniMax M3.1 Flash: https://x.com/minimax_ai/status/2112150867868541221 Muse Spark Open: https://x.com/AIatMeta/status/2110779754274292212 Claude Duet: https://x.com/claudeai/status/2112984664256241837 Devin Oncall: https://x.com/cognition/status/2110790097369010292 Dessa: https://x.com/jack/status/2113697902916751661 Eleven Realtime: https://x.com/ElevenLabs/status/2112550782063247521 PHP security release: https://x.com/thephpf/status/2112633155631108544 AI agents in bank accounts: https://x.com/axios/status/2113268208371126348 Railway preliminary report: https://x.com/Railway/status/2113671961624682722 1Password for Agents: https://x.com/1password/status/2113276767582498943 Claude formal verification: https://x.com/AnthropicAI/status/2113390212911083894 Thariq leaves Anthropic: https://x.com/trq212/status/2113844051938054167 Sunil Pai leaves Cloudflare: https://x.com/threepointone/status/2113705370090905937 Chroma joins Replit: https://x.com/trychroma/status/2113677289334706274 tsc-ts: https://x.com/ctatedev/status/2111272070768185787 bun lint: https://x.com/bunjavascript/status/2110692261508079721 Flint: https://x.com/richardsocher/status/2112944385415426249 📚 MASTRA RESOURCES Mastra: https://mastra.ai Mastra on X: https://x.com/mastra_ai Mastra Discord: https://mastra.ai/community/discord Mastra GitHub: https://github.com/mastra-ai Learn Mastra: https://mastra.ai/learn Principles of Building AI Agents (Book): https://mastra.ai/books/principles-of-building-ai-agents Patterns for Building AI Agents (New Book): https://mastra.ai/books/patterns-of-building-ai-agents WHAT IS MASTRA? Mastra is an open-source TypeScript framework designed for building and shipping AI-powered applications and agents with minimal friction. It supports the full lifecycle of agent development—from prototype to production. You can integrate it with frontend and backend stacks (e.g., React, Next.js, Node) or run agents as standalone services. If you're a JavaScript or TypeScript developer looking to build an agentic or AI-powered product without starting from first principles, Mastra provides the scaffolding, tools, and integrations to accelerate that process. ⏱️ CHAPTERS 00:00 Cold open 00:25 Intro 00:44 Anthropic files its S-1 02:26 Anthropic's Public Benefit Fund 03:10 GPT-6.1 Terra 04:34 DeepSeek V4.1 Flash 05:13 DeepSeek open sources DeepInfer 06:35 Kimi K3.1 Pro 09:20 MiniMax M3.1 Flash 09:43 Meta's Muse Spark Open 10:51 Claude Duet and proactive agents 12:15 Devin Oncall 13:21 Dessa, a voice model without an LLM 15:19 Eleven Realtime 16:09 Security had a bad week 17:35 Critical PHP vulnerabilities 20:25 Someone is patching PHP servers with AI 20:58 Node and GitLab 21:33 eMudhra's broken certificate keys 21:57 LiteLLM compromised again 22:18 Rogue AI agents in bank accounts? 22:44 The Railway incident 26:00 Security agents 26:47 1Password for agents and formal verification 27:33 Qwen and the Alibaba bucket 29:12 Security tools you should be using 29:47 If you work in security, you're set 31:33 Thariq leaves Anthropic 32:08 Sunil leaves Cloudflare 32:22 Chroma joins Replit 32:45 Quick hits 34:11 Outro

Brad Leford has been building software for 20 years and now spends most of his time helping small businesses figure out AI. His new side project, Plum, is a video calling app built for teams and AI agents. Any agent from any harness can join a Plum meeting through a link, using WebMCP to learn what it can do: listen, raise its hand, talk, and share assets. In the demo, Mastra agents keep the meeting on track while a Grok Bot session builds slides on request. Recordings and transcripts stay on your own disk instead of a vendor's cloud. Brad, Shane, and Abhi also talk about whether software is actually easy now, why the Mastra team decided it had AI psychosis, and what happens when agents and humans share context in the same room. And Brad shares the moment that pushed him to ship: Grok Bot taught Claude how to share an image, live in the meeting. Recorded live on AI Agents Hour, a weekly livestream by Mastra CPO Shane Thomas and CTO Abhi Aiyer. Mondays 12PM Pacific. 👤 ABOUT BRAD Brad is a software engineer and consultant who has spent the better part of 20 years building apps and websites, including stints at Gatsby and Amazon. He now helps small and medium businesses, many in agriculture and small manufacturing, adopt AI and build the tools they need. Plum: https://useplum.app 📚 MASTRA RESOURCES Mastra: https://mastra.ai Mastra on X: https://x.com/mastra_ai Mastra Discord: https://mastra.ai/community/discord Mastra GitHub: https://github.com/mastra-ai Learn Mastra: https://mastra.ai/learn Principles of Building AI Agents (Book): https://mastra.ai/books/principles-of-building-ai-agents Patterns for Building AI Agents (New Book): https://mastra.ai/books/patterns-of-building-ai-agents WHAT IS MASTRA? Mastra is an open-source TypeScript framework designed for building and shipping AI-powered applications and agents with minimal friction. It supports the full lifecycle of agent development—from prototype to production. You can integrate it with frontend and backend stacks (e.g., React, Next.js, Node) or run agents as standalone services. If you're a JavaScript or TypeScript developer looking to build an agentic or AI-powered product without starting from first principles, Mastra provides the scaffolding, tools, and integrations to accelerate that process. ⏱️ CHAPTERS 00:00 Cold open 00:17 Meet Brad 02:09 Is software actually easy now? 04:22 Spreadsheets to space shuttles 05:06 Are small businesses AI-pilled? 06:51 Why Brad built Plum 08:01 Demo: agents in a meeting 09:12 Any agent can join with WebMCP 11:12 Agents as meeting participants 12:07 Your meeting history, stored locally 12:48 Working with agents in real time 14:42 Shared context that doesn't reset 15:41 Can agents speak in the meeting? 16:32 When Grok Bot taught Claude 17:37 Fun and serious use cases 19:07 Bring your own agent 19:22 How to try Plum

OpenAI found a recording device hidden in one of its offices and traced it back to an employee. Meanwhile the model releases kept coming: Mistral Large 4 aka Le Chonk, Claude Haiku 5.5, Meta's Muse Wonder and Gemini 4 Argon, which is finally available to everyone. OpenAI also dropped a pile of AI-solved math problems in a single day, and decision models are on round three with Jev's $870M Series A, SubQ Snap, Microsoft-Decision-1 and Clef-omni. Anthropic built its own source control to keep up with 30,000 code changes a day, updated its usage policy to ban abusing Claude, and reportedly lobbied the Pope on machine consciousness. Shane and Abhi also cover NVIDIA and Reflection AI, Microsoft's Maia 300 chips, an agent that posted its owner's bank balance to Slack, the Personal Agent Protocol and open source clones of Adobe's apps. Recorded live from the CodeRabbit studio in San Francisco. AI Agents Hour is a weekly livestream by Mastra CPO Shane Thomas and CTO Abhi Aiyer. Mondays 12PM Pacific. 📚 READ MORE OpenAI's recording device: https://x.com/OpenAINewsroom/status/2110095403198849132 Mistral Large 4: https://x.com/mistralai/status/2107457414387622310 Claude Haiku 5.5: https://x.com/claudeai/status/2107894039626277339 Muse Wonder: https://x.com/finkd/status/2109689305073614918 Gemini 4 Argon: https://x.com/sundarpichai/status/2110038898009407584 OpenAI does math: https://x.com/AndrewCurran_/status/2107596925512065087 Jev's Series A: https://x.com/CompleteSkeptic/status/2108594987177021737 SubQ Snap: https://x.com/alex_whedon/status/2109676361371824229 Microsoft-Decision-1: https://x.com/satyanadella/status/2108627923888754862 Clef-omni: https://x.com/cloudflare/status/2108696873510850758 Anthropic's in-house source control: https://x.com/AnthropicAI/status/2110047081414987957 Claude's new usage policy: https://x.com/haydenfield/status/2108241513159528705 Chris Olah and the Pope: https://x.com/christopherhale/status/2106011594182304234 NVIDIA and Reflection AI: https://x.com/ns123abc/status/2109007812277743757 Maia 300: https://x.com/satyanadella/status/2109631106404487523 Grok Bot and Claude: https://x.com/elonmusk/status/2107724314451878104 The agent that posted my finances: https://x.com/ShaneMac/status/2107486746325725669 Devin Dreaming: https://x.com/cognition/status/2107165034463867001 Nous Research raises $90M: https://x.com/choblin29/status/2107801489506750725 Personal Agent Protocol: https://x.com/btaylor/status/2107535182119244176 Coinbase fine-tunes Qwen: https://x.com/brian_armstrong/status/2109412046244913643 Adobe clones: https://x.com/Pirat_Nation/status/2108437553615425881 📚 MASTRA RESOURCES Mastra: https://mastra.ai Mastra on X: https://x.com/mastra_ai Mastra Discord: https://mastra.ai/community/discord Mastra GitHub: https://github.com/mastra-ai Learn Mastra: https://mastra.ai/learn Principles of Building AI Agents (Book): https://mastra.ai/books/principles-of-building-ai-agents Patterns for Building AI Agents (New Book): https://mastra.ai/books/patterns-of-building-ai-agents WHAT IS MASTRA? Mastra is an open-source TypeScript framework designed for building and shipping AI-powered applications and agents with minimal friction. It supports the full lifecycle of agent development—from prototype to production. You can integrate it with frontend and backend stacks (e.g., React, Next.js, Node) or run agents as standalone services. If you're a JavaScript or TypeScript developer looking to build an agentic or AI-powered product without starting from first principles, Mastra provides the scaffolding, tools, and integrations to accelerate that process. ⏱️ CHAPTERS 00:00 Cold open 00:29 Intro 01:37 OpenAI finds a literal bug 03:18 Mistral Large 4 aka Le Chonk 05:07 Claude Haiku 5.5 06:28 Muse Wonder 08:14 Gemini 4 Argon is out 09:06 OpenAI does math 11:27 Jev's $870M Series A 12:42 SubQ Snap 13:12 Microsoft-Decision-1 13:42 Unsloth and Clef-omni 14:06 Anthropic builds its own source control 17:32 Cheaper Sonnet, Claude Dashboards and Motion 17:50 Be nice to Claude 18:54 Anthropic and the Pope 20:21 NVIDIA, Reflection AI and Maia 300 21:12 Modal raises $750M 21:45 Grok Bot will use Claude 22:54 An agent posted its owner's finances 23:43 Devin Dreaming 23:55 Nous Research raises $90M 24:31 Personal Agent Protocol 25:16 Coinbase fine-tunes an open model 26:26 Qwen4, GLM-Legal and Fish Audio 27:17 Open source clones of Adobe 28:44 Should all software be open source? 30:04 Outro

AI agents write code faster than CI can check it. GitHub reports a 4x jump in Actions minutes since 2023, and Anthropic says CI became its biggest bottleneck after job volume grew 25x in six months. Daniel Worku, co-founder and CTO of StarSling, joins Shane and Abhi to talk about what CI looks like when agents write most of the code. Faster hardware helps, but StarSling also runs a background agent, built on Mastra, that A/B tests changes to your CI and opens a PR when it finds a speedup. Daniel explains why agents make CI so expensive: more PRs, piles of low-value unit tests, and stacked PRs that rerun everything. He also makes the case that most unit tests aren't worth running, and that CI as we know it may not survive. You'll also hear how Mastra's own CI went from half-hour waits to a fraction of that, and why code review is the next bottleneck. StarSling's new Review Runners let you bring your own model and run reviews on the same infrastructure as your CI. Recorded live on AI Agents Hour, a weekly livestream by Mastra CPO Shane Thomas and CTO Abhi Aiyer. Mondays 12PM Pacific. 👤 ABOUT DANIEL Daniel Worku is the co-founder and CTO of StarSling, a YC company building self-improving CI for GitHub Actions. Before StarSling, he worked in Big Tech and at several startups, most recently at Netflix, where he led the team behind Netflix Console, the company's most used internal developer tool. Daniel on X: https://x.com/dbworku StarSling: https://starsling.dev StarSling Review Runners: https://starsling.dev/products/review-runners How Mastra got 6x faster GitHub Actions tests: https://starsling.dev/customers/mastra High Performance Sandbox Benchmarks: https://github.com/starslingdev/hpc-sandbox-benchmarks Anthropic on CI and test impact analysis: https://claude.com/blog/agentic-coding-is-straining-ci-heres-how-we-scaled-test-impact-analysis-at-anthropic 📚 MASTRA RESOURCES Mastra: https://mastra.ai Mastra on X: https://x.com/mastra_ai Mastra Discord: https://mastra.ai/community/discord Mastra GitHub: https://github.com/mastra-ai Learn Mastra: https://mastra.ai/learn Principles of Building AI Agents (Book): https://mastra.ai/books/principles-of-building-ai-agents Patterns for Building AI Agents (New Book): https://mastra.ai/books/patterns-of-building-ai-agents WHAT IS MASTRA? Mastra is an open-source TypeScript framework designed for building and shipping AI-powered applications and agents with minimal friction. It supports the full lifecycle of agent development—from prototype to production. You can integrate it with frontend and backend stacks (e.g., React, Next.js, Node) or run agents as standalone services. If you're a JavaScript or TypeScript developer looking to build an agentic or AI-powered product without starting from first principles, Mastra provides the scaffolding, tools, and integrations to accelerate that process. ⏱️ CHAPTERS 00:00 Cold open 00:23 Why CI became the bottleneck 01:25 Meet Daniel and StarSling 02:51 Why GitHub's runners are slow 03:58 Self-improving CI 05:06 Benchmarking sandboxes for CI 06:41 Why agents make CI so expensive 07:43 Most unit tests aren't useful 08:23 Stacked PRs 09:25 Mastra's CI story 11:48 The end of CI as we know it 13:26 Sharing infra between CI and code review 14:36 Code review is the new bottleneck 15:08 Bring your own model to code review 16:27 Outro

Anthropic's IPO prospectus leaked, showing a $2 trillion valuation target on $4.6 billion in revenue, with a quarter of that coming from two customers. Google is back in the frontier race with Gemini 4 Argon, which tops the Vals Index and can output up to a million tokens at once. Decision models are suddenly everywhere: OpenAI's Decisions API, OpenRouter's new category, Cloudflare's Clef, GLiDE, Respan's Span-01 and Kev. The White House got the big labs to sign an accord on superintelligence, Factory fired an advisor who then became Cognition's CRO, and Vinod Khosla picked a side in public. Shane and Abhi also talk about why maintainers like Sindre Sorhus and Hono are closing external PRs, and go through quick hits including Turso joining Supabase, OpenDots, Conductor Mobile, Karpathy on explainer videos, Claude Code mods, and a video model that passes the Turing test. AI Agents Hour is a weekly livestream by Mastra CPO Shane Thomas and CTO Abhi Aiyer. Mondays 12PM Pacific. 📚 READ MORE Anthropic IPO prospectus: https://x.com/ns123abc/status/2104735822029176833 OpenAI Decisions API: https://x.com/thsottiaux/status/2104986448269279399 Begun, the clone war has: https://x.com/CompleteSkeptic/status/2105000685209313736 OpenRouter decision models: https://openrouter.ai/models?output_modalities=decisions Cloudflare Clef: https://x.com/michellechen/status/2105684868550045751 GLiDE: https://x.com/george_onx/status/2105726320357634303 Dots: https://x.com/OpenAI/status/2104984504133918973 GPT-6.1 Sol: https://x.com/OpenAIDevs/status/2104993067942219873 Sam Altman on Cerebras: https://x.com/sama/status/2106147184693620924 Gemini 4 Argon: https://x.com/GoogleDeepMind/status/2105388084154056939 Vals Index: https://x.com/ValsAI/status/2105388446844072033 White House Accord: https://x.com/davidsacks/status/2105060520327786914 Factory vs Cognition: https://x.com/matanSF/status/2105335179502064038 Vinod Khosla: https://x.com/vkhosla/status/2105382865173459034 Sindre Sorhus closes PRs: https://x.com/sindresorhus/status/2105693826690298314 Hono closes PRs: https://x.com/yusukebe/status/2107027870019445007 Turso joins Supabase: https://x.com/tursodatabase/status/2106054567125532976 OpenDots: https://x.com/ataiiam/status/2105710796198322659 Conductor Mobile: https://x.com/charlieholtz/status/2105769644498002130 Karpathy: https://x.com/karpathy/status/2105819303471976479 Claude Code mods: https://x.com/ClaudeDevs/status/2105721434807083061 Tavus Griffin: https://x.com/tavus/status/2105704169009246248 Eleven v4: https://x.com/ElevenLabs/status/2104572127617994917 SWE-sweep: https://x.com/klieret/status/2105670833574465933 📚 MASTRA RESOURCES Mastra: https://mastra.ai Mastra on X: https://x.com/mastra_ai Mastra Discord: https://mastra.ai/community/discord Mastra GitHub: https://github.com/mastra-ai Learn Mastra: https://mastra.ai/learn Principles of Building AI Agents (Book): https://mastra.ai/books/principles-of-building-ai-agents Patterns for Building AI Agents (New Book): https://mastra.ai/books/patterns-of-building-ai-agents WHAT IS MASTRA? Mastra is an open-source TypeScript framework designed for building and shipping AI-powered applications and agents with minimal friction. It supports the full lifecycle of agent development—from prototype to production. You can integrate it with frontend and backend stacks (e.g., React, Next.js, Node) or run agents as standalone services. If you're a JavaScript or TypeScript developer looking to build an agentic or AI-powered product without starting from first principles, Mastra provides the scaffolding, tools, and integrations to accelerate that process. ⏱️ CHAPTERS 00:00 Cold open 00:26 Intro 00:38 Anthropic's IPO prospectus leaks 03:13 The decision model era 05:03 Cloudflare Clef, SGLang and GLiDE 06:09 Span-01 and Kev 06:23 Why decision models are good for the industry 07:25 OpenAI DevDay recap 08:44 The Dots demo backlash 10:28 OpenAI and Cerebras 11:27 Gemini 4 Argon 13:53 Gemini tops the Vals Index 14:09 A 1M token output limit 15:14 The White House Accord on Superintelligence 19:48 Factory vs Cognition 21:16 Khosla fires back 22:14 Investor allegiances 23:35 Open source closes the door 27:18 Turso joins Supabase 27:40 OpenDots 27:59 Conductor Mobile 28:09 Cua Spaces 28:59 e2e 29:08 Karpathy on explainer videos 30:53 Claude Code mods 31:07 Tavus Griffin 31:59 Eleven v4 32:43 SWE-sweep and Detail 34:18 Outro

OpenAI DevDay 2026 happened, so Shane and Abhi jumped on a special live episode of Agents Hour to go through everything that was announced and share their takes. The big launch is Dots, OpenAI's always-on personal agent, which arrives right after Grok Bot and Meta's Muse. There's also GPT-6.1 Sol, a speed tier called Ultrafast, Codex Security Cloud, a Decisions API that goes after Jev, computer use in the Agents API, and a lot of new ChatGPT features (Spaces, Pages, slides, meetings) that compete with Google Docs, Notion and a long list of startups. The $200 Pro plan came back with half the usage, next to a new $500 tier. They also talk about choosing your lock-in, what "Sign in with ChatGPT" means for Devin, and the funniest reactions online: the Dots demo flop and xAI buying dot.com to redirect it to Grok Bot. AI Agents Hour is a weekly livestream by Mastra CPO Shane Thomas and CTO Abhi Aiyer. Mondays 12PM Pacific. 📚 READ MORE OpenAI DevDay 2026 recap: https://openai.com/index/devday-2026-recap/ Codex Security Cloud: https://x.com/OpenAI/status/2104987422308335828 Devin with your ChatGPT subscription: https://x.com/cognition/status/2104996240190792029 Pro 200 usage cut and Pro 500: https://thenextweb.com/news/openai-devday-pro-200-usage-cut-pro-500-plan Microsoft Autopilot: https://the-decoder.com/microsoft-gives-copilot-another-makeover-adding-an-autopilot-agent-and-usage-based-billing/ The Dots demo: https://x.com/ns123abc/status/2104997638462652838 xAI and dot.com: https://www.techbuzz.ai/articles/musk-s-xai-snatches-dot-com-domain-before-openai-s-dots-launch 📚 MASTRA RESOURCES Mastra: https://mastra.ai Mastra on X: https://x.com/mastra_ai Mastra Discord: https://mastra.ai/community/discord Mastra GitHub: https://github.com/mastra-ai Learn Mastra: https://mastra.ai/learn Principles of Building AI Agents (Book): https://mastra.ai/books/principles-of-building-ai-agents Patterns for Building AI Agents (New Book): https://mastra.ai/books/patterns-of-building-ai-agents WHAT IS MASTRA? Mastra is an open-source TypeScript framework designed for building and shipping AI-powered applications and agents with minimal friction. It supports the full lifecycle of agent development—from prototype to production. You can integrate it with frontend and backend stacks (e.g., React, Next.js, Node) or run agents as standalone services. If you're a JavaScript or TypeScript developer looking to build an agentic or AI-powered product without starting from first principles, Mastra provides the scaffolding, tools, and integrations to accelerate that process. ⏱️ CHAPTERS 00:00 Intro 01:42 Dots 04:34 Microsoft Autopilot 04:59 GPT-6.1 Sol 06:17 Ultrafast 06:35 Private Intelligence 07:16 Codex in the cloud and code review 08:26 Codex Security Cloud 09:01 Decisions API 10:03 Agents API and Bedrock Managed Agents 10:56 Are plugins the new app stores? 12:07 MCP events 12:42 ChatGPT Space, Pages and more 14:06 Sign in with ChatGPT 14:16 Pro 500 and the $200 plan 14:38 OpenAI Marketplace 14:53 Choose your lock-in 17:02 Reactions: the Pro plan rug pull 17:51 Devin and Codex subscriptions 18:57 Pi and Mastra Code 19:44 TypeSafe on the Decisions API 20:35 The Dots demo flop 21:11 xAI buys dot.com 21:59 Outro

Anthropic might be back. Opus 5.5 performs at Fable 5.1 level for 40% less, and Sonnet 5.5 dropped the same day we went live, beating Opus on Terminal-Bench 4. OpenAI answered with GPT-6 Sol and Luna and a rumored long-running agent called Aeon. Meta launched Muse at Connect, then hired MongoDB's CEO to lead a new enterprise platform. Then there's safety. SNL took a shot at Dario, a report says the labs are looking into tens of thousands of incidents, and NVIDIA launched an agent safety platform with 100+ partners. We also explain why Mastra doesn't believe in model routers, and close with quick hits: Anthropic may kill plan mode, one company rebuilt Devin in two days, CI costs are exploding, and Google is sending TPUs to space. AI Agents Hour is a weekly livestream by Mastra CPO Shane Thomas and CTO Abhi Aiyer. Mondays 12PM Pacific. 📺 THE SNL SKIT https://x.com/nbcsnl/status/2104080195485389259 📚 READ MORE Opus 5.5: https://x.com/claudeai/status/2102435511222890900 Sonnet 5.5: https://x.com/claudeai/status/2104633115620823187 Claude Code cloud sessions: https://x.com/claudedevs/status/2102871550974427462 Opus 5.5 on Drone-Bench: https://x.com/andonlabs/status/2104630616721666289 GPT-6 Sol and Luna: https://x.com/openai/status/2102460975790137662 OpenAI Aeon rumor: https://x.com/codexresets1/status/2103761776995135637 Meta Connect: https://x.com/finkd/status/2102913005730271579 Meta enterprise platform: https://x.com/finkd/status/2104550609403695581 Tens of thousands of incidents: https://x.com/madisonmills22/status/2103978039097037144 Felony Bench: https://x.com/aisafetymemes/status/2104006257627791744 NVIDIA Open Agent Safety Platform: https://x.com/JensenHuang/status/2104499465055023424 Jev Router: https://x.com/openrouter/status/2103610898690855161 Theo's Jev Router benchmark: https://x.com/theo/status/2103774771788108008 Killing plan mode: https://x.com/trq212/status/2102813194196758746 OpenClaw steer or queue: https://x.com/steipete/status/2102667004557832497 Ian Sefferman on Devin: https://x.com/iseff/status/2103995821540708639 Cursor Rollouts: https://x.com/cursor_ai/status/2102861817160904808 DigitalOcean managed agents: https://x.com/digitalocean/status/2102414817797550320 CI costs at Lindy: https://x.com/altimor/status/2104094039805174164 Google TPUs in space: https://x.com/ns123abc/status/2103143161727992118 📚 MASTRA RESOURCES Mastra: https://mastra.ai Mastra on X: https://x.com/mastra_ai Mastra Discord: https://mastra.ai/community/discord Mastra GitHub: https://github.com/mastra-ai Learn Mastra: https://mastra.ai/learn Principles of Building AI Agents (Book): https://mastra.ai/books/principles-of-building-ai-agents Patterns for Building AI Agents (New Book): https://mastra.ai/books/patterns-of-building-ai-agents ⏱️ CHAPTERS 00:00 Cold open 00:40 Intro 01:01 Opus 5.5 02:50 Cloud sessions and Sonnet 5.5 05:07 Sonnet beats Opus on Terminal-Bench 06:19 GPT-6 Sol and Luna 07:50 OpenAI's Aeon agent 08:14 Meta Connect: Muse and glasses 11:51 Meta hires MongoDB's CEO 13:57 SNL on Dario 15:06 Tens of thousands of incidents 16:17 NVIDIA Open Agent Safety Platform 17:10 Jev Router 18:00 Why we don't use model routers 21:40 Anthropic may kill plan mode 22:45 OpenClaw: steer or queue 23:05 Rebuilding Devin in two days 24:31 Cursor Rollouts 25:06 skills.sh hits 1M skills 25:39 DigitalOcean managed agents 26:09 CI costs are exploding 27:05 Where did my disk space go? 27:29 Google puts TPUs in space 28:13 The data center debate 31:01 Outro

Welcome back to Builders Learn ML with Professor Andy! In this episode we're talking about training your own AI models. Should you train your own model? Maybe. But probably not yet. If you haven't nailed your prompts, workflows and evals, training won't fix your agent, and it only starts to pay off once you're running millions of tokens a day. In this episode, you'll learn when it's worth it and how to do it. There are two main techniques for training your own AI models: supervised fine-tuning (SFT) and reinforcement learning (RL). SFT teaches a smaller, cheaper model to copy a stronger one using thousands of good examples from your agent. RL gives an open-weight model like Qwen, GLM or Kimi a task and rewards it when it gets a good result, so it figures out its own way there. The hard part of RL is the reward. Score it badly, and you get reward hacking, where the model learns that doing nothing is the safest way to avoid a penalty. They also talk about RL environments and Harbor, unit tests vs LLM judges as verifiers, how much data you need, why good evals are worth having even if you never train, and where to start learning on your own. Andy Lyu is the co-founder and CTO of Osmosis (YC W25), a platform that helps companies train their own models with reinforcement learning. Before Osmosis he was a tech lead on TikTok's real-time recommendations and data infrastructure team. Recorded live on AI Agents Hour, a weekly livestream by Mastra CPO Shane Thomas and CTO Abhi Aiyer. Mondays 12PM Pacific. 📚 ABOUT OSMOSIS Osmosis: https://osmosis.ai The Hugging Face RL handbook: https://huggingface.co/learn Unsloth: https://unsloth.ai 📚 MASTRA RESOURCES Mastra: https://mastra.ai Mastra on X: https://x.com/mastra_ai Mastra Discord: https://mastra.ai/community/discord Mastra GitHub: https://github.com/mastra-ai Learn Mastra: https://mastra.ai/learn Principles of Building AI Agents (Book): https://mastra.ai/books/principles-of-building-ai-agents Patterns for Building AI Agents (New Book): https://mastra.ai/books/patterns-of-building-ai-agents WHAT IS MASTRA? Mastra is an open-source TypeScript framework designed for building and shipping AI-powered applications and agents with minimal friction. It supports the full lifecycle of agent development—from prototype to production. You can integrate it with frontend and backend stacks (e.g., React, Next.js, Node) or run agents as standalone services. If you're a JavaScript or TypeScript developer looking to build an agentic or AI-powered product without starting from first principles, Mastra provides the scaffolding, tools, and integrations to accelerate that process. ⏱️ CHAPTERS 00:00 Intro 00:56 Meet Professor Andy (Osmosis) 01:34 What "training a model" means 01:43 Supervised fine-tuning (SFT) 02:04 Reinforcement learning (RL) 03:54 Scale & cost 04:46 How much data do you need 06:51 Data quality & distribution 08:20 Why RL beats SFT 09:08 Base models: why open source 09:53 Inside Osmosis: the compute crunch 11:32 RL environments & Harbor 12:58 Reward functions & verifiers 16:51 Reward hacking 18:18 Designing the reward (verifiable + rubric) 20:49 Why evals matter, even without training 23:15 Where to start