OpenAI’s platform chief on Codex, agents, and the next AI hiring shift
Published 2026-02-12 - Runtime about 80 min - Watch on YouTube
TLDR
- Sherwin Wu says 95% of OpenAI engineers use Codex daily, and 100% of PRs get Codex review.
- OpenAI sees engineers shifting from writing code to managing 10-20 parallel agent threads, with 70% more PRs from heavier Codex users.
Key Takeaways
- OpenAI is treating Codex as a second pair of eyes, cutting review work from 10-15 minutes to 2-3 minutes.
- Wu thinks the next 12-24 months are a rare window for engineers before the job fully transforms.
- Managers may become higher leverage too, using ChatGPT with GitHub, Notion, and Google Docs to review performance and context.
- OpenAI’s platform strategy is explicit: the API, Responses, Agents SDK, AgentKit, and eval tools are stacked to help others build.
Notes
- Wu leads engineering for OpenAI’s API and developer platform, where Codex is deeply embedded in daily engineering workflows.
- He says most engineers at OpenAI now use Codex every day, and Codex reviews every PR that reaches production.
- He frames the role shift as IC engineers becoming tech leads who manage fleets of agents rather than typing every line themselves.
- Some engineers are juggling 10 to 20 parallel Codex threads at once, steering, checking, and correcting them.
- He cites a 70% increase in PRs among heavier Codex users, and says the gap keeps widening over time.
- OpenAI is running an internal experiment on a 100% Codex-written codebase, exposing the limits of fully automated development.
- Wu says the hardest part is not generating code but keeping agents on track when they drift or need correction.
- He compares the moment to SICP’s wizard metaphor and the Sorcerer’s Apprentice: powerful incantations, but only if you know how to direct them.
- For managers, AI is widening the spread between top performers and everyone else, so he spends more time with the strongest people.
- He expects managers to become higher leverage by using AI for research, organizational context, and performance review synthesis.
- OpenAI’s platform philosophy is ecosystem-first: every model shipped in products also lands in the API, and competitors are not blocked.
- Wu says startups should focus on customer resonance, not fear of OpenAI crushing them; the opportunity space is large enough for many winners.