Database › OpenAI
OpenAI
No formal AI-use mandate has been published. OpenAI describes near-total internal use of its own Codex agent, with engineers moving most of their work to it by December 2025.[2]
Tools
Metrics tracked
- Codex accounts for 99.8% of weekly output tokens inside OpenAI; the average engineer now generates 99% of output tokens with Codex, and engineering usage is 27x its November 2025 level.[2]
- Codex writes nearly 80% of OpenAI's code, according to its India strategy head.[1]
- Codex writes more than 90% of the Codex app's own code; engineers typically run 4-8 agents in parallel.[3]
Guardrails
- A bespoke AI code-review model runs on every PR; about 9 in 10 of its comments are judged valid. Non-critical code can merge after AI review alone, while core agent and open-source code still needs a human reviewer.[3]
Hiring, reviews & process changes
- The team says the traditional PR flow is 'starting to crack' under the volume of agent-written code.[3]
- Legal, finance and recruiting staff switched to agent-based work around April 2026.[2]
Reported results
- The heaviest (99th-percentile) users run more than 60 hours of agent turns per day across parallel agents.[2]
Metric tags: % code AI-generated, Adoption rate, Agent autonomy, Code review, AI-authored code
Editor's note: Vendor-authored: OpenAI sells Codex and is reporting on its own use of it. openai.com returns a bot challenge to automated checks but loads for readers. Size band is an editorial estimate.
Sources
- Outlook Business: OpenAI's strategy head in India reveals Codex writes 80% of company's code · 2026-05-15 · press source
- OpenAI: How agents are transforming work · 2026-06-25 · vendor source
- The Pragmatic Engineer: How Codex is built · 2026-02-17 · press source
Last verified 2026-10-25. Industry and size are editorial bands; see methodology. Spot an error? Submit a correction.