TL;DR
Altman called India the image-gen hotspot, while Garry Tan said AI teams will eat incumbents instead of trimming costs. Peter Yang slammed fake-benchmark eval theater, and Aaron Levie, Aditya Agarwal, and Every all pushed the same line: AI can write, but humans still own meaning.
ChatGPT Images 2.0 is already seeing massive traction in India, with more than 1 billion images created there. The takeaway: image generation isn’t a novelty anymore — it’s becoming a real consumer habit at scale.
He says the real move isn’t squeezing expenses — it’s building a cracked AI-human symbiosis team that can outgrow giant incumbents. He also name-drops ZeroEntropy as his top pick for personal AI search against his 120k-markdown-brain setup.
He says coding is getting abundant, but that doesn’t mean technical talent matters less. The real shift is that engineers, designers, bankers, and analysts who know their craft can steer AI agents, catch mistakes, and ship way more than novices. He also thinks colleges and companies need to rethink pipelines fast, because demand for technical people is spreading across every industry.
Build evals from real customer traces and feedback, not generic academic benchmarks. The point is to keep signal as models get smarter — and to use actual conversations to build product sense, with Claude helping synthesize the themes.
People overrate the difficulty of making something and underrate the grind of getting anyone to care. It’s a blunt reminder for builders: shipping is only half the game — distribution is where most products die.
He says employees joining recently minted unicorns need to do the math: tranched valuations can leave you with a much higher strike price, a painful 409a, and equity that looks juicier than it really is. His advice is simple — use Claude or ChatGPT to model the exit, and join for the learning and mission, not just the headline comp.
He says AI helped him write a lot of code on Friday, but watching the FA Cup Final and his kid’s dance recital on Saturday was the reminder: humans aren’t running out of purpose. It’s a simple, optimistic take from a former Dropbox CTO and Facebook engineer — AI can take more work, but not the parts of life that actually matter.
Claude Code works best as a thinking partner, not a writer
The Takeaway: Claude Code becomes far more useful when it helps you think, not when it tries to draft.
Noah Breyer, now running Alethic, an AI strategy consultancy, has turned his Obsidian vault into a living second brain powered by Claude Code. He starts Claude in the root of his entire note system, so it can search across years of markdown files, chats, PDFs, and project folders. For a new talk, he’ll seed the vault with context, tell the model he’s “in thinking mode, not writing mode,” and let it pull relevant material from his archive before he ever touches a draft.
That distinction matters to him. “I think partially because we call it generative, there’s entirely too much focus on its ability to write and not enough focus on its ability to read,” he says. His workflow is built around reading, connecting, and interrogating ideas. He even uses a sub-agent as a collaborative interviewer: “Do not under any circumstances want you to try to write it. Take this literally.”
The other big shift is portability. By rigging a home server and connecting it to his phone, Noah can do serious work in the car, on the move, or between meetings. He uses voice mode for long, exploratory sessions on topics like self-attention, Walter Benjamin, and “transformers eating the world.” The point isn’t convenience for its own sake; it’s that deep thinking no longer has to wait for a desk.
Peter Steinberger called AI code review the killer dev workflow, while Thariq said Claude Code turned files into a workbench. Cursor bet on specialized models plus brutal RL infra, and Anthropic detailed how it contained Claude across products.
Steinberger said AI agents need lean skills, not bloated prose; Levie argued AI won’t kill jobs, it’ll raise the bar. Yang said Codex tested itself while Claude still owned frontend, and Garry Tan pushed hard evals as the way agents got better fast.
Levie warned AI demos fool CEOs; the hard work starts after. Tan said high agency plus taste beat everything, while Yang urged builders to build the system that builds the MVP first. LeCun kept the pressure on: LLMs help, but they’re not the road to real intelligence.
Levie said AI kills tasks, not jobs, so roles just expand. Masad said Replit’s AI stack can ship MVPs in a weekend, while Tan showed Qwen fine-tunes in hours and personal AI got real. No Priors said Cerebras won by betting on speed before the market cared.
Levie said AI will surface more bugs, making humans the bottleneck. Steinberger argued GitHub’s native PR limits kill bot policing, while Tan pushed the “ship the 60%” rule and Swyx said AI coding should harden codebases, not just move faster.