
The Changing Role of Code Review

Building Codex with Tibo Sottiaux
Topic 14 of 20
Published Sep 9, 2026 · Clip 46:40–52:08 (5:28)
Tibo describes developing a code review model capable of spotting deep logic and reasoning mistakes, now superhuman in benchmarks and part of mainline models, with security review mandatory and blocking PRs across OpenAI. He argues code review was always about correctness plus a social ritual for information exchange; the correctness and security parts are being automated, while discussions about intent move outside the PR. He frames the future as agreeing on a box and its invariants, with freedom to change anything inside without discussion, preserving attention.