Follow builders, not influencers.
TL;DR
Rauch said teams were building their own design factories, while Steinberger called open-source AI security a full-time arms race. Masad priced OSS trust in compute, and Woodward shipped Gemini on Mac in 100 days.
He says the future of design isn’t one killer SaaS tool — it’s every team spinning up a custom internal "design factory" for exactly their needs. His example: Shader Lab, built with Claude Code, Three.js, Next.js, and Vercel for Basement Studio, is a sign that assembling software from building blocks is getting easier than buying the right product.
He says the last 4 months turned OpenClaw into a much harder target: sandboxing, allow-lists, and per-access exec prompts, all hammered on by hundreds of security researchers. His bigger point is blunt — if you’re shipping open-source AI harnesses and not publishing advisories, you’re leaving users blind while attackers keep iterating.
He argues that if frontier models can fully automate vuln hunting, open-source packages need a new trust signal: a visible metric for how much compute was spent securing them. In his example, a repo like linux would show a huge “security spend” score — basically turning hardening into a public badge, not a hidden chore.
He says the real edge isn’t just AI speed — it’s the long grind to clarity, depth, and persistence that machines still don’t have. Even if Claude can one-shot a system diagram, getting to that level of precision still takes a human who cares.
They shipped Gemini on Mac: a fully native Swift app with 100+ features built by a small team in under 100 days. The pitch is speed and polish — lightning-fast, 100% native, and clearly aimed at making Gemini feel like a real desktop product, not just a web wrapper.
Cursor is adding /canvas, a way to build custom interfaces for agents instead of forcing everything through a chat box. It’s a small-looking feature with a big implication: agent workflows are moving from text-only prompts to actual product surfaces.
AI speeds up one part of the process, then the constraint pops up somewhere else, so total demand for humans can actually rise. He points to legal, security, healthcare, sales, and even new company formation as places where automation creates more downstream work — not less. The real risk, he says, is standing still while competitors use AI to outbuild, out-sell, and out-serve you.
He says every product builder should visit China at least once, because the country’s AI work culture, EV ecosystem, and absurdly cheap delivery economy are hard to grasp from the outside. The takeaway is less travelogue, more warning: if you build products, you need to understand how fast China moves and how different the operating model is.
Claude Code’s 1M context window is a double-edged sword: it unlocks bigger tasks, but it also makes context pollution easier if you don’t manage sessions well. He points to autocompact controls, including setting your own threshold, so power users can effectively dial the window down — e.g. 400k — when that’s the better tradeoff.
He says Meta’s AI story has moved from the Llama 4 flop to the “turn”: more hiring, Zuck coding again, an Opus-ish model GA, and acquisitions like Dreamer and Manus to build an AI OS/prosumer layer. He also argues subagents are just optimization — the real leap is agents that compose and manage other agents, which is why Cognition’s new Spaces launch matters.
She argues that in AI, “play” isn’t fluff — it’s the job: poking at models non-utilitarianly is often where the best discoveries come from. She also showed off Frontend Slides in @AnyGenIO, saying once you make slides with HTML, it’s hard to go back.
Verifiable AI beats token guessing for mission-critical work
The Takeaway: Eve is betting that the next useful AI won’t be a better guesser — it’ll be a model you can inspect, verify, and trust.
Eve, founder and CEO of Logical Intelligence, is building around a blunt premise: if AI is going to touch software, chips, cars, or anything else where failure matters, “correctness” can’t be an afterthought. Her company works on both LLMs and EBMs, but the long-term bet is on energy-based reasoning models with latent variables — a mouthful she also calls Kona. Her argument is that LLMs are black boxes that only let you judge the final answer, while EBMs let you see the model’s internal state as it learns.
She keeps coming back to a simple contrast: language models are forced to translate everything into tokens, even when the task is spatial, physical, or mechanical. That’s why she says using LLMs for things like driving or circuit control is like using “the literature department everywhere.” EBMs, by contrast, build an energy landscape and search for the lowest, most probable state. In her couch example, the model isn’t guessing the next word — it’s learning the most likely configuration.
Her core philosophy is practical, not mystical: if you need milliseconds, sparse data, or verifiable output, don’t ask a language model to pretend it’s a physics engine. “We don’t have to,” she says. “There are EBMs.”
Woodward said Gemini’s turning into a test-prep machine, Albert called Claude Code the whole workspace, and Cat Wu shipped a desktop control center with parallel sessions and review tools. Rauch also argued agent builders need elastic Postgres, not vibes.
Rauch said the moat moved from code to the code factory, while Levie argued every team now needed an agent wrangler. Cursor leaned into customizable multi-agent views, Replit added region controls, and No Priors backed Periodic Labs’ bet that AI could learn atoms by running experiments.
Amjad Masad said Apple’s 50th has turned into a PR disaster, while Aaron Levie argued agents would create more work, not cut jobs. Rauch pushed engineers into the customer hot seat, and Claude warned teams to harden security fast.
Thariq said Claude Code now handles TurboTax pain, while Rauch called microVM sandboxes the new compute layer. Aditya Agarwal pushed memory over loops, and Levie argued AI won’t shrink law—it’ll inflate it.
Claude pushed into Word with tracked edits, and Claude Code moved planning to the web with auto mode approvals. Garry Tan called agents the Altair BASIC era, while Aaron Levie warned software without a real API gets left behind.
Karpathy said free ChatGPT lagged while frontier coding models didn’t. Albert pushed cheap-to-smart escalation, Rauch said cloud infra went agent-native, and OpenAI’s next leap looked like autonomy—not chat.
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.
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.