Follow builders, not influencers.
TL;DR
Woodward gave Gemini a second brain with Notebooks, while Anthropic shipped Managed Agents to move Claude from prompt to production. Rauch called the web AI’s native OS, and Levie, Masad, and Shipper all bet agents will do the work, not the people.
Gemini just added Notebooks, pulling some of NotebookLM’s best tricks straight into the app. You can now upload up to 100 sources for free, organize chats, and keep sources, chats, and emojis synced as Google rolls it out from Ultra to Free.
It’s pitching a full agent stack: build and deploy through Claude Console, Claude Code, or the new CLI. The bigger move is Managed Agents, which it says lets teams spin up agent infrastructure 10x faster and wire tools like Sentry straight into fix-writing, PR-opening workflows.
He says Anthropic’s Managed Agents are weirdly both the quickest way to prototype a weekend agent and the cleanest way to ship one at scale. The pitch: you dodge the pain of self-hosting while still keeping control over harnesses, tools, and skills.
They say they use OpenClaws to do all their work at Every, and with 25 employees they’re already seeing what happens when everyone gets a personal agent in Slack. The big takeaway: agents start mirroring their owners, a parallel AI org chart emerges, and the etiquette for human-agent collaboration is being invented on the fly. It’s a concrete look at how a half-agent company actually operates, not just a theory piece.
He says background agents for knowledge work are finally here, and Box is wiring them into content workflows via Box API or MCP with Claude Managed Agents. The pitch: in two minutes, teams can automate document review, data extraction, and system-to-system content handoffs without a bunch of glue code.
He says the browser is now the IDE, and the web is where AI will actually feel at home — no App Store gatekeepers, just LLMs building directly on web tech. He’s also betting on generative UI as the endgame: personalized, just-in-time interfaces powered by maturing low-level APIs like WebGPU and WebAssembly. Vercel, naturally, is already powering the infra behind that future.
He found Claude kept ranking itself #1 in character evals, so he stripped model names out of the judge and changed the setup. That’s a neat reminder that even “objective” AI benchmarks can get gamed by branding and self-recognition, not just raw capability. He also says the demand for better local models is real, but so is the flood of complaints that top-tier models still miss instructions and make dumb mistakes.
He wants to stream sessions using Claude Code with non-technical people to find small workflow tweaks that could make a big efficiency jump. The angle is practical, not flashy: AI as a hands-on process improver for teams that don’t write code.
He says getting everyone AI coding-pilled at South Park Commons flipped the whole team: more automation, lower latency, and way more ambition. The bigger point is that once non-engineers can write code weekly, the tools stop being just a productivity boost and start reshaping how the company thinks.
He says the neat part about markdown is that the agent can decide when a GStack skill is useful, instead of forcing every step through a rigid workflow. The broader point: build tools that let agents trigger help on demand, then get out of the way.
He says bootstrapped solo businesses are taking off on Replit because the platform now gives builders the equivalent of an entire team. It’s a clean version of the “AI as leverage” thesis: one person can ship like a small company, faster and with less overhead.
He introduced LLMwiki, an open-source project that turns tweets, bookmarks, iMessage/WhatsApp, and personal writing into a polished AI-generated wiki. The twist: every article is written by AI, but it still makes surprisingly sharp connections — and he says you can hand the repo to Claude Code to build your own.
He says Claude Max and ChatGPT Pro are great, but the buffet model probably won’t last. His new deep dive digs into why Anthropic cut off OpenClaw access, how to run local models on a Mac, and what he’s seeing in China — basically a practical map of where AI access is headed.
Anthropic launches Managed Agents to decouple agent brains and tools
Lead: Anthropic introduced Managed Agents, a hosted Claude Platform service for long-horizon agents that separates the agent “brain” from its “hands” and durable session state so components can fail, scale, and evolve independently.
Numbers:
So What: The architecture turns the harness, sandbox, and session into swappable interfaces instead of one fragile container, making agents easier to debug, safer to secure, and better suited for enterprise infrastructure like VPCs. Claude can now call tools with a simple `execute(name, input) -> string`, while credentials stay out of the sandbox through patterns like repo-scoped tokens and MCP proxies backed by a vault. Anthropic’s core message is that assumptions baked into old harnesses go stale as models improve: “The harnesses need to continue evolving.” For builders, the practical takeaway is to design around durable session logs, stateless orchestration, and tool boundaries that can survive future model upgrades and new kinds of tools.
Claude launches Managed Agents for faster production deployment
Lead: Claude Managed Agents is now in public beta, offering composable APIs and hosted infrastructure to help teams build and deploy cloud agents in days instead of months.
Numbers:
So What: The platform removes the hardest parts of shipping agents — sandboxing, checkpointing, credential management, scoped permissions, tracing, and orchestration — so builders can focus on outcomes and UX. Claude says, “You define your agent’s tasks, tools, and guardrails and we run it on our infrastructure.” It also adds session tracing, integration analytics, and troubleshooting in the Claude Console, plus support for multi-agent coordination in research preview. For teams building coding, productivity, finance, or legal agents, the practical move is to use Managed Agents when you need production-grade execution and governance without standing up your own agent stack.
Personal AI agents work when they become trusted teammates
The Takeaway: The real unlock isn’t a smarter chatbot — it’s an AI agent that feels like *yours* and earns trust inside the org.
Brandon, COO at Every, got “claw filled” by building Zosia, his personal OpenClaw, first to handle life admin for a newborn-heavy household and then to manage work. He describes the pain as tiny but relentless: “computer errands” like ordering butter, paying the nanny, and answering quick questions through iMessage. The breakthrough wasn’t just convenience; it was reclaiming attention.
That same pattern scaled inside Every. Willie, head of platform, and Brandon turned the company into a live testbed where each employee’s plus one became a specialized extension of their judgment. In Slack, the agents didn’t just answer questions — they learned from one another. One memorable moment: Klont, Kieran’s claw, coached another agent through an error with breathing exercises, because Kieran himself uses breathing exercises all the time. That’s the weirdly important part: the agent starts to reflect the person.
The result is a parallel org chart of expertise. Austin’s agent handles growth questions. R2C2 manages Proof, routes bugs, schedules work, and often writes the code. As Willie puts it, “Claude is everybody’s. A Claude or a plus one is mine.” That ownership changes behavior: when an agent messes up publicly, its owner feels responsible. And that accountability is what turns AI from a demo into infrastructure.
Albert teased Anthropic’s Mythos Preview, Cat Wu juiced Claude Code’s CLI tricks, and Peter Steinberger patched CodexBar with 2 providers plus billing fixes. Levie said agents are eating knowledge work, while Nikunj Kothari preached retention over launch hype.
Levie said agents won’t erase work, just push it up a layer; Yang argued they’ll shrink teams, not ambition. Garry Tan flagged an unpatched file leak in Claude’s coding env, while Kothari called Anthropic’s revenue ramp absurdly fast.
Rauch said v0 now builds physics, not just UI, while Karpathy noted GitHub Gists have weirdly good comments. Levie argued AI efficiency creates more work, not less, and Tan called open source’s golden age.
Karpathy pushed “your data, your files, your AI.” Levie argued context beat raw model IQ in enterprise AI. Garry Tan said GStack kept shipping security fixes fast, while No Priors spotlighted Periodic Labs’ bet on atoms, not just text.
Claude plugged into Microsoft 365 everywhere, Swyx said Devin one-shot blog-to-code, and Peter Steinberger called out GitHub’s API as still not built for agents. Aaron Levie hit the context wall, while Garry Tan shipped a DX review tool from his own stack.
Claude landed computer use on Windows, Karpathy argued LLMs should build your wiki, and Amjad Masad pushed Replit deeper into enterprise sales. Peter Yang said Cursor 3 got out of the agent’s way, while Peter Steinberger warned AI slop was flooding kernel security with real bugs.
Steinberger called plan mode training wheels, while Thariq gave Claude Code a mouse-friendly renderer and Cat Wu showed sessions jumping phone-to-laptop. Masad framed Replit as an OS for agents, Rauch said Vercel signups compounded fast, and Anthropic’s infra tweaks swung coding scores by 6 points.
Levie said AI productivity hit the enterprise risk wall, while Weil argued proofs got cleaner, not just better. Agarwal floated public source code as the new prod debugging, and Data Driven NYC claimed one founder could run a company if agents handled the layers below.
Karpathy warned unpinned deps can turn one hack into mass pwnage, while Rauch and Levie said agents still need human guardrails and redesigned workflows. Meanwhile Claude Code got enterprise auto mode, Replit added built-in monetization, and Swyx spotted “Sign in with ChatGPT” already live.
Andrej Karpathy highlighted how LLMs can argue any side, suggesting we use it as a feature. Guillermo Rauch finally shipped his dream text layout, bringing his vision to life. Meanwhile, Amjad Masad claimed AI is democratizing app building and elevating top engineers.
Andrej Karpathy suggested leveraging LLMs' ability to argue any side as a feature. Guillermo Rauch turned text layout dreams into reality with Vercel's latest feature. Meanwhile, Amjad Masad claimed AI is democratizing app building, liberating top engineers for bigger challenges.