Sherwin Wu V2

Sherwin Wu, Head of Engineering for OpenAI's API and Developer Platform, discusses how AI coding agents are transforming engineering roles, management, and startup economics. The conversation covers practical Codex workflows, the bitter lesson of building scaffolding around fast-improving models, business process automation as an overlooked opportunity, and OpenAI's platform philosophy.

01Key learnings

• Build for where models are going rather than where they are today, since products that are almost-working now can become excellent as capabilities improve.
• Treat fast-changing AI scaffolding skeptically: models often absorb tooling like vector stores and agent frameworks, so avoid over-investing in layers they will eat.
• Don't blindly follow customer feature requests in AI; customers may anchor on local maxima while the underlying models are shifting.
• Encode tribal knowledge into repositories through docs, AGENTS.md, Skills, and code comments so coding agents can perform tasks without human context.
• Spend more than half your time with top performers, since AI tools amplify high-agency people and widen productivity gaps across teams.
• For AI adoption inside companies, combine executive buy-in with a bottom-up team of enthusiastic, technically adjacent people who build best practices and evangelize them.
• Don't over-worry about platform owners squashing your startup; durable businesses come from products customers truly love, and OpenAI emphasizes staying a neutral ecosystem platform.
• Look beyond software engineering: repeatable business processes across industries are a large, underrated opportunity for AI automation.

02Frameworks mentioned

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.