OpenAI Codex lead details iterative refinement cycle

Thibault Sottiaux, Engineering Lead on OpenAI's Codex team, described the repeated cycle of shipping updates, collecting feedback, fixing issues, and occasionally resetting before resuming work. Harrison Chase and engineer Viv outlined a comparable loop for AI agents that begins with an initial shipped version after internal dogfooding, then gathers tracing data, applies agentic compute to surface failures, creates evaluations, and makes targeted edits before repeating the sequence.

Original post

TI

Tibo#768@THSOTTIAUX

Ship => Feedback => Ship => Feedback => Ship => Oops something not quite right => Reset => Yay => Ship => … The never ending cycle of Codex improvements

2:18 PM · May 16, 2026 View on X