Follow builders, not influencers.
TL;DR
Claude plugged into Microsoft 365, Devin turned blog posts into working code, and Garry Tan shipped a DX review tool. Peter Steinberger said GitHub’s API still wasn’t built for agents, while Aaron Levie warned agents hit the same wall as managers.
Microsoft 365 connectors are now on every Claude plan, so Outlook, OneDrive, and SharePoint can feed straight into chats. It’s a practical move: Claude is turning from a standalone assistant into something that can actually work across your inbox and docs.
He says he keeps slamming into GitHub API quota limits, which is a pretty blunt sign the platform still isn’t designed for agentic workflows. In a separate thread, he also argues Anthropic’s latest move hurts the ecosystem, though he credits Boris for trying to soften the blow with cache fixes that should lower API costs.
He says a $10B company suing because an app “looks and feels” like theirs is exactly the wrong move. Instead of lawyering up, he’s pushing Bevel Health as the free alternative to WHOOP and framing it as a mission to make great health accessible to everyone.
He says AI agents don’t escape the old management problem: humans can only hold so much context, so delegation exists for a reason. Until agents can self-track, escalate cleanly, and plug into a broader system, they’ll still need real human oversight — which is why the “jobs disappear” take is overblown.
He released /plan-devex-review on GStack, saying he builds tools for himself first and is now sharing it to help teams create better developer experiences. He also pointed to Addy Osmani’s DX framework as the inspiration, which makes this feel like a practical, opinionated workflow tool rather than another generic AI wrapper.
He argues that when design gets jammed into product, explorations get shallow and brand suffers — better to keep brand/product design close, but not subordinate. He also says PMM belongs with product management, while design + PM creates bad incentives that blur the work.
He says he’s hitting a weird new threshold: paste in blog posts and tweets, and Devin one-shots the implementation. The punchline is the model underneath — Gemini Flash Lite — was way out of distribution for this task, but it still just worked, which is a pretty strong signal for agentic coding workflows.
AI’s boom is real, but the buildout will be messy
The Takeaway: AI isn’t a hype cycle so much as an 80-year backlog finally cashing out.
Marc Andreessen, cofounder of Andreessen Horowitz, frames AI as the culmination of decades of serious research, not a sudden miracle. He calls it an “eighty year overnight success,” arguing that today’s leaps are the payoff from work that started with the 1943 neural network paper and ran through the Dartmouth era, the 1980s expert-systems boom, AlexNet, and transformers. His point is blunt: “now we know the neural network is the correct architecture,” and the recent jump from chatbots to reasoning models to coding agents is what finally made the tech feel real in the world.
What changed his mind from cautious optimism to full conviction was the sequence of functional breakthroughs. LLMs were impressive, but reasoning models like o1 and r1 answered the skepticism about whether these systems could do more than autocomplete. Then coding crossed the line — if AI can hold up against serious programmers, it can spread everywhere else. From there, agents and self-improvement make the trajectory feel less like a demo and more like a platform shift.
Still, Andreessen is not naive about the downside. He warns that the dot-com crash wasn’t just a demand story; it was an overbuilt infrastructure story. The same mistake could happen again if capital floods into data centers and GPUs faster than real demand justifies. But unlike the old telecom boom, today’s builders are Microsoft, Amazon, Google, Meta, Nvidia, OpenAI, and Anthropic — companies with real cash, real revenue, and real capacity pressure. That makes the upside look durable, even if the path there gets ugly.
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.