Topics›Autonomous Agents with Reliable Outcomes›OpenAI Codex Tech Lead: Spec → Prompt → 6 Right-Sized PRs → CI Over Local Tests
Very HighAct: HighFull content
OpenAI Codex Tech Lead: Spec → Prompt → 6 Right-Sized PRs → CI Over Local Tests
Engineering Leadership Gregor·June 4, 2026
Reading List
Summary
Michael Bolin (OpenAI Codex tech lead) describes his personal AI-assisted engineering workflow: starts with a written spec, writes a prompt from the spec, Codex generates 6 PRs for a single large feature (right-sized for review), relies on CI over local testing, has a 'babysitting PRs/CI' skill for monitoring. Key judgment: 'the spec is where the human intelligence is concentrated — the prompt converts that into agent instructions.'
Why this signal
Neo's #3 focus area. The 6-PR decomposition for a big feature is a specific, novel technique for managing AI-generated code review load at team scale. Directly applicable to Thomas Engineering's Claude Code workflows.
Original email
Engineering Leadership Gregor · 6/4/2026
How OpenAI Codex Tech Lead Does AI-Assisted Engineering