Bio
Head of Engineering for the OpenAI API and Developer Platform. Runs the team that ships Codex internally. Direct counterpart to Cat Wu (Anthropic) and Boris Cherny in the cohort of operators publishing on the operating model behind frontier-lab product velocity.
Operating themes
- The bottleneck is context, not capability. Encode tribal knowledge.
- Build for the next model. Current-model scaffolding will be eaten.
- Top performers benefit disproportionately. Spread widens with AI tools.
- ICs become tech leads of agent fleets. The role shape is shifting.
Cards
- When the agent isn't doing what you want, fix the context — not the model — Fix the context, not the model, when the agent fails [Tier A]
- Build for the model six months out — the current model will eat your scaffolding — Build for the model six months out [Tier A]
- AI tools widen the spread between top and bottom performers — invest in top performers — Invest in top performers; the spread is widening [Tier A]
Sources captured
- 2026-04-28 — Lenny's Podcast, "Codex inside OpenAI, engineers as managers" (
raw/podcasts/sherwin-wu--openai-codex-engineers-as-managers--2026-04-28.md)