Skip to main content
aifeed.dev the frontpage of AI
0

How OpenAI's Codex Team Actually Works

Thibault Sottiaux, engineering lead on the Codex team, walks through how a roughly 40-person group ships one of OpenAI's core products. The setup is deliberately flat - most projects run with two or three people who own everything end-to-end, meetings are rare, and ideas flow bottom-up. What stands out is how aggressively they use their own tool internally: a single PM triaged over 100 bugs in an hour using Codex, new engineers onboard through the product itself, and large Python-to-Rust rewrites happen with AI assistance. Hundreds of reusable prompt skills encode team knowledge in a way that traditional documentation never managed to. The pattern of AI teams dogfooding their own coding tools is becoming the clearest signal of which products are actually production-ready.

// 0 comments

> login to comment