Back from San Francisco (S10E02) - Sous les briques, le soleilBack from San Francisco (S10E02)<br>May 28, 2026 10:00 AM<br>Back from San Francisco. S10E02.<br>S10E01 was the robotaxis. S10E02 is the coffees.<br>Between two redwood contemplations, I had a lot of coffees with old contacts: former founders, Epitech, Telecom, and LIP6 people.<br>That was the point of the trip: take the pulse from senior SWEs, SREs, CTOs, founders, and adjust how I think about angel investing.<br>This is what unsettled me most.<br>Everyone wants to work at Anthropic or OpenAI.<br>The talent pool that would have considered Google or Meta 10 years ago now points at frontier labs. Experienced founders and CTOs I could easily imagine starting again are in hiring loops there.<br>This is the main difference with Paris.<br>Will that still be true post-IPO?<br>Senior SWEs are now product engineers.<br>They serve product, growth, marketing, whatever needs shipping.<br>You need to be good at code and architecture. You also need taste. And people skills.<br>Product differentiation is accelerating. Tech no longer differentiates enough. Product does.<br>Paris has taste. That could put us in a good place.<br>CTOs are coding again. A lot.<br>The code inflation question is real. Full-LLM codebases will raise hard questions. LLMs love adding code. Refactoring is harder.<br>That is a question for the future. The present is already busy enough.<br>Founders sound resigned about lab velocity.<br>This was a few weeks after Claude Design shipped and Codex accelerated. The common belief: if the TAM is above $1B, the labs will eat it.<br>Building DevTools means competing with Claude Code and Codex. Hard.<br>The better move is to use them as leverage.<br>That is starting.<br>Agentic companies are tired and excited.<br>New models ship every quarter from frontier labs and China. Every release forces teams to reassess agent quality.<br>A prompt that worked on Opus 4.6 will not produce the same result on 4.7. Even less on 5.5. And then Qwen enters the room.<br>Lots of work ahead for evals.<br>Token cost arbitrage is now a CFO topic.<br>Why let a frontier model handle prompts that could maybe run on a small model in a closet?<br>Silicon Valley is in a rupture period.<br>&ldquo;The labs take everything&rdquo; is one version of the story. The other is more practical: the quality of iteration outside pure tech is accelerating.<br>The opportunities to accelerate are exceptional.