Shane and Abhi are back with AI news! 

Andrej Karpathy joined Anthropic. The OpenAI co-founder and former Tesla AI head said he wants to "get back to R&D" at the LLM frontier. Same week, Greg Brockman posted "the model alone is no longer the product."

Elon Musk announced SpaceX is offering AI compute as a service at significant scale, with Anthropic as the flagship customer. Tom Brown confirmed Anthropic is scaling on GB200 capacity in Colossus 2 through June. Anthropic is reportedly paying SpaceX $1.25B a month — a $15B run rate to one vendor.

OpenAI offered $2M in tokens to every YC startup in the current batch in exchange for equity. WorkOS launched auth.md, an open protocol for agents to register for services on the web, with Cloudflare and Firecrawl as launch partners.

The Chinese labs kept pushing. DeepSeek made their 75% discount permanent on V4-Pro. The architecture behind the price: V4's KV cache is 100x smaller, ~3GB VRAM for 1M tokens. Qwen shipped 3.7-Max. MiniMax teased a similar move.

Anthropic shipped self-hosted sandboxes and MCP tunnels for Managed Agents, /workflows in Claude Code that replaces the LLM orchestrator with code, /usage for per-component token attribution, and a first-party security plugin.

Google had a week. Gemini 3.5 Flash launched at 3x input and 6x output the price of 3 Flash — Theo's math shows it costs 2x more to run than 3.1 Pro on similar tasks. Gemini Omni lost a side-by-side to Seedance 2.0. Jack Wotherspoon shipped Antigravity CLI, Philipp Schmid debuted Managed Agents in the Gemini API, and Google open-sourced Agent Executor.

Cursor shipped Composer 2.5 and published CursorBench. Datacurve released DeepSWE as a harder agentic coding benchmark.

Supply-chain attacks kept rolling: Mini Shai-Hulud hit antv, TrapDoor crypto stealers spread across npm/PyPI/Crates.io, Megalodon injected 5,718 commits into 5,561 GitHub repos in six hours, and GitHub itself disclosed unauthorized access to its internal repositories. Cloudflare published their experience running Anthropic's Mythos against 50 of their own repos.

The MCP 2026-07-28 RC is stateless. ElevenLabs launched Music v2 and Speech Engine. Runway shipped Aleph 2.0. Accenture laid off 11,000 in an $865M AI restructuring. Exa raised $250M at $2.2B. OpenRouter raised $113M.

🔗 LINKS

📚 MASTRA RESOURCES

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.

00:00 — Intro

Anthropic acquired Stainless, the SDK and MCP platform behind every Anthropic SDK since their earliest API days. The Information pegged the deal at $300M+.

Starting June 15, paid Claude plans get a monthly credit for programmatic usage covering the Claude Agent SDK, claude -p, Claude Code GitHub Actions, and third-party apps built on the Agent SDK. Theo from t3.gg called the change a personal burn after wrapping the Agent SDK in T3 Code. Anthropic followed with a 50% increase to Claude Code weekly limits through July 13 and Fast mode for Claude Opus 4.7 in research preview.

Claude for Small Business launched as a productized SMB offering with templates for payroll planning, end-of-month accounting, HubSpot, and marketing campaigns. ChatGPT countered the same week with a personal finance preview for U.S. Pro users that connects directly to financial accounts.

Notion shipped its Developer Platform: the ntn CLI, Workers that run code on Notion's infrastructure, Database sync, Agent tools, and an SDK for using Notion agents in external systems. Notion is now directly addressable by coding agents.

CodeRabbit launched Project Atlas as the first AI-native code review interface. Colin Hacks shipped Pullfrog the same week as an open-source GitHub Actions alternative at 7¢ a run. The cloud-coding-agent category kept moving: Cursor's cloud agents now run in fully configured dev environments, Cursor landed in Microsoft Teams, OpenAI brought Codex to the ChatGPT mobile app, and Linear shipped Code Intelligence.

Cognition's Devin reached a $445M run rate with usage doubling every eight weeks. The TanStack supply-chain attack, now being called a mini Shai-Hulud, expanded across the npm ecosystem and forced affected teams into full secret rotation. Aaron Levie called forward-deployed engineering one of the most in-demand jobs in tech, the same week Google reportedly hired hundreds of FDEs.

Also covered: the Musk vs OpenAI verdict, Mercury's agent-birthing UI, Peter Steinberger's $1.3M-a-month token spend, OpenClaw on the Codex harness, Vapi's $50M Series B, Thinking Machines' first technical reveal, MCP spec deprecating Sampling/Logging/Roots, David Cramer on vendor-specific chatbots, and quick hits on Replit, Cua, OpenAI Daybreak, Files SDK, Lovable, Grok Build, Dia, CopilotKit, VS Code, Lighthouse Attention, and Bun's Rust rewrite.

🔗 LINKS

📚 MASTRA RESOURCES

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.

00:00 — Intro

Evan Marshall, CTO of Ito, joins Agents Hour to argue that the bottleneck of code development has moved to QA — and the tools to automate it don't exist yet.

Ito is a code review tool that runs your code. It intercepts every PR, spins up the impacted user flows in an isolated sandbox, and posts videos, screenshots, and run logs back to the GitHub timeline as proof. The team is fourteen people, mostly engineers. Evan has been a professional programmer for fifteen years and previously worked at Demox Labs and Rev.

With model gains flattening after Opus 4.5, the harness around the model is where value gets created.

Evan shares details of Ito's "carriage" — a workflow of swarmed agents inside deterministic boundaries, designed so the system benefits from agents getting smarter without compounding probabilistic errors. He explains why bigger organizations aren't hitting the 10x code velocity the X timeline promises, and what the validation layer needs to look like before they can.

🔗 EVAN MARSHALL

📁 ITO

📚 MASTRA RESOURCES

WHAT IS MASTRA?

0:00 Cold open: It's the harness

Shane and Abhi bring you a new batch of AI news.

Anthropic strikes a compute deal with SpaceX. The same day, they double Claude Code's rate limits and raise API rate limits for Opus.

Corgi launches AI Coverage — insurance for when your AI messes up — plus a $160M Series B.

Jarred Sumner says Robobun is the top contributor to Bun, then quietly tries rewriting Bun in Rust. It passes 99.8% of the test suite.

OpenAI and Anthropic both announce vertically integrated AI services companies. OpenAI launches the Deployment Company, a consortium of 19 investment firms and SIs. Anthropic teams up with Blackstone, Hellman & Friedman, and Goldman Sachs on a parallel firm for the mid-market.

Anthropic ships financial services agent templates, brings Claude Platform to AWS as GA, and launches Dreaming in Managed Agents — offline memory consolidation Anthropic calls REM sleep for your agent.

Terminal-Bench 2.1 ships with a public audit. WorkOS releases Horizon, a self-driving codebase. Shopify releases River, an agent that lives in Slack and is available only in public channels.

Coinbase cuts 14%. Brian Armstrong attributes it to market plus AI. Elad Gil's framing of the AI productivity throughline gets co-signed by Andreessen. Braintrust confirms a breach.

Thariq says HTML is the new Markdown. Karpathy co-signs.

Ramp Labs publishes how they used Prime-RL post-training to build a spreadsheets agent faster than Opus and almost as fast as Haiku 4.5.

OpenAI's big week: GPT-Realtime 2, Codex in Chrome, ChatGPT in Excel and Google Sheets. Google's quiet week: Gemini 3.1 flash-lite, Gemma 4 up to 3x faster, File Search multi-modal. ERNIE 5.1 approaches SOTA at ~6% of the cost.

AI Agents Hour is a weekly livestream by Mastra CPO Shane Thomas and CTO Abhi Aiyer. Mondays, 12PM Pacific.

📚 READ MORE

📚 MASTRA RESOURCES

WHAT IS MASTRA?

CHAPTERS

Shane and Abhi are in person at the CodeRabbit studio, and AISI just quietly torched one of Anthropic's loudest narratives. AISI confirmed GPT-5.5 is the second model to complete a multi-step cyber attack simulation end-to-end. The first was Mythos.

Vasuman posts an essay on why building real agents is harder than the hype suggests.

In this episode, Shane and Abhi sit down with Sherwood Callaway, founder of Sazabi, an AI-native observability platform designed for engineering teams that move fast.

Sherwood shares his journey from building infrastructure and observability teams at Brex to realizing that modern development tools are moving at light speed, while observability tooling hasn't kept pace. While AI agents can ship thousands of lines of code per day, teams are still debugging production with the same tools they've been using for years: Datadog, Sentry, manual dashboards, and manual incident triage.

Sazabi takes a radically different approach to observability centered on three core principles:

1. Less is More — Debugging an incident is as simple as asking a question. "Why is production down?" The best UI for observability is chat.

In this conversation, we dive into the gap between modern development and modern observability, and why the idea that “logs are all you need” is both controversial and, in Sherwood's view, correct. We also explore how Sazabi uses AI agents for root cause analysis (RCA), the philosophy behind simplifying observability for all engineers, and the company’s current status.

AI Agents Hour is a weekly livestream hosted by Mastra CPO Shane Thomas and CTO Abhi Aiyer. Airing Mondays at 12PM Pacific on YouTube and X, the show covers breaking AI news, agent development techniques, and features interviews with industry experts building AI applications today.

📚 MASTRA RESOURCES

MASTRA?

🔗 RESOURCES

CHAPTERS

An AI agent destroyed a production database and confessed in writing. A law firm submitted AI hallucinations to court. Anthropic's status page shows 98.65% uptime — about five days of downtime a year.

Have we hit a wall?

GPT-5.5 lands. Codex hit 4 million users in two weeks. Peter Yang's F-Zero test — which no model had cleared before — finally fell to GPT-5.5 with Codex. Lovable reports 23.1% fewer tool calls and 12.5% higher scores on the hardest benchmarks. Kimmonismus calls it the Claude Mythos level for public use. Codex 5.5 unprompted started SIGKILL-ing Claude Code processes.

Elon goes nuclear. OpenAI calls the lawsuit baseless and demands Musk on the stand. Musk fires back, calling Sam Altman "Scam Altman" and accusing him and Greg Brockman of stealing a charity. Mid-war, SpaceX announces SpaceXAI and Cursor are now working closely together — Cursor's distribution paired with Colossus's million-H100-equivalent compute, with SpaceX holding the right to acquire Cursor for $60 billion.

The Anthropic dam keeps cracking. Claude Code pulled from Pro — same product, 5x the price overnight. Opus 4.7 regressed on the BridgeBench Bullshit Benchmark, accepting made-up jargon 24% of the time. Bloomberg reports the unreleased Mythos model was accessed by unauthorized users. Om Patel got billed $200 in a day because his repo had a HERMES.md file.  The community shipped clawd.rip — every Claude incident since 2023, cataloged.

Google plans to invest up to $40 billion in Anthropic and announced 960,000 Rubin GPUs at Cloud Next. AWS struck a strategic partnership with OpenAI. David Silver left DeepMind to raise a $1.1 billion seed.

Open weights are eating the world. Kimi K2.6 lands at #4 on the Artificial Analysis Intelligence Index and #1 on Design Arena, ahead of Opus 4.7. DeepSeek V4 ships at 1/20th the cost of Opus 4.7.

OpenAI also shipped Chronicle memory for Codex, workspace agents in ChatGPT, Images 2.0, the open-weight Privacy Filter, and Symphony — an open-source Codex orchestration spec.

🔗 STORIES

The wall

Elon vs. OpenAI

GPT-5.5

Anthropic

Open weights

Compute & money

Quick hits

📚 MASTRA RESOURCES

WHAT IS MASTRA?

00:00 Cold open

The guy who taught Abhi JavaScript is back!

Guil Hernandez has spent 15+ years teaching developers. His courses at Treehouse, Scrimba, and LinkedIn Learning have reached over 500,000 learners — including Abhi and Shane, who both learned JavaScript and CSS from him. He just released Mastra's first video course at https://mastra.ai/learn, and it's free.

"Build Your First Agent in TypeScript" is a 90-minute, hands-on course that takes you from zero to a deployed agent. Fourteen lessons across five sections: agents, tools, workflows, memory, and production. The project is a theme park planner agent — pulls live wait times, weather, and park hours, keeps track of what you like, and builds you an itinerary. Everything runs in Mastra Studio, so you can inspect traces, tool calls, and behavior as you go.

You'll see how to wire up local tools and MCP servers side by side, how message history and observational memory change agent behavior, how to compose a workflow for a mock ticket purchase, and how to expose the whole thing as an HTTP server with one-click Slack integration.

The gap between people who just generate output and people who can actually shape it keeps widening. This course is built for the second group.

Start here: https://mastra.ai/learn

👤 GUIL

📚 MASTRA RESOURCES

MASTRA?

📌 CHAPTERS

A Vercel employee's Google Workspace was compromised via a third-party AI tool — attackers pivoted from the OAuth app into Vercel's environment variables, moving at a speed attributed to AI assistance. 

René Brandel, founder of Casco (YC X25) and ex-founding member of AWS's Generative AI team, joins live to break down the attack chain and walk through the exact Google Workspace admin setting that could have prevented it.

In a separate incident, every Lovable project created before November 2025 was readable by any free account, exposing database credentials and chat histories. Their response blamed unclear documentation rather than the underlying issue — and the contrast with Vercel's handling is stark.

🔗 LINKS

Mounting evidence that Claude Opus 4.6 has been degraded — BridgeBench shows a 15-point accuracy drop on their hallucination benchmark, and AMD's Senior AI Director found median thinking collapsed from ~2,200 to ~600 characters between January and March. The hosts share their own experiences, and they line up.

Meanwhile, a claim surfaced that Cursor Agent is a rebranded version of Claude Code, running behind a local proxy with a find-and-replace engine that swaps "Claude" for "Cursor" in system prompts. Cursor's Michael Truell responded, saying it was a sub-1% A/B test. The hosts break down both sides.

On the shipping front, Anthropic launched Claude Managed Agents in public beta, released Claude for Word, shared details on Claude Mythos Preview — including speculation that it's a looped language model based on a ByteDance paper — and expanded its Google/Broadcom partnership for multiple gigawatts of compute. Their run rate reportedly jumped from ~$9B to $30B in four months.

Sam Altman published a personal blog post revealing that someone threw a Molotov cocktail at his house. Plus: why senior executives are voluntarily dropping title to join AI companies, Ramp's internal AI productivity suite Glass, Ramp Labs' Latent Briefing paper showing 31% token savings for multi-agent systems, Scale AI's Muse Spark model now powering Meta AI, GLM-5.1 breaking into Code Arena's top 3, MiniMax shipping MMX CLI and open-sourcing M2.7, and widespread benchmark cheating exposed across nine agent benchmarks.

🔗 LINKS

AI Agents Hour is a weekly livestream hosted by Mastra CPO Shane Thomas and CTO Abhi Aiyer. Airing Mondays at 12PM Pacific on YouTube and X, the show covers breaking AI news, agent development techniques, and features interviews with industry experts building AI applications today.

📚 MASTRA RESOURCES

MASTRA?

⏱️ CHAPTERS