💻 openai/codex: How OpenAI scales agentic development
Engineer John Wang (CTO of Assembled) analyzed openai/codex as a verifiable 'ground truth' for agentic development: in May 2025, the Rust portion received 98 commits from 6 authors, while in the first 25 days of August 2026 — over 1,000 commits from 135 authors with a team of 137 people.
🌍 Agentic development best practices here can be verified through open data: OpenAI turns code review comments into AGENTS.md rules (322 lines), then into lint checks — a total of 38 rules. Tests — 615,000 lines, ~40% of the codebase.
👤 Patterns are transferable for free: AGENTS.md with prohibitions on bypassing sandbox checks and 'tests for the sake of tests', lint for ambiguous arguments, tests for the affected crate during development. The repository (~121,000 stars) and Wang's analysis — a checklist for teams with coding agents.
Source 1: https://johnjwang.com/post/2026/08/27/learnings-from-the-codex-repo Source 2: https://github.com/openai/codex