How does OpenAI's Codex AI build itself?

Codex uses itself to write research harnesses for its own training and processes user feedback to decide on future development.

What is the 'AI teammate' concept at OpenAI?

OpenAI engineers treat Codex as a teammate, assigning it tasks via project management tools and collaborating on code.

How fast was the Sora Android app built with Codex?

The Sora Android app was built from scratch in 18 days using Codex, demonstrating rapid development capabilities.

Home / Technology / OpenAI's AI Codes Itself: A Recursive Revolution

OpenAI's AI Codes Itself: A Recursive Revolution

13 Dec, 2025

•

Summary

OpenAI's Codex AI agent now largely builds and improves itself.
The Sora Android app was developed from scratch in 18 days.
Codex is treated as an 'AI teammate,' integrating with workflows.

OpenAI's AI Codes Itself: A Recursive Revolution

OpenAI's powerful AI coding agent, Codex, is now largely self-building and self-improving, marking a significant leap in artificial intelligence development. This recursive process allows the AI to enhance its own capabilities, accelerating innovation at an unprecedented pace. The development of the Sora Android app in a mere 18 days exemplifies this efficiency.

Codex is increasingly integrated into OpenAI's engineering practices, functioning as a virtual "AI teammate." It assists with everything from planning architecture to fixing bugs, seamlessly fitting into project management tools like Linear and communication platforms like Slack. This allows engineers to assign tasks directly to Codex, fostering a collaborative environment.

This approach positions Codex as a junior developer, amplifying human potential rather than replacing it. OpenAI envisions extending this capability beyond software engineers to individuals without prior programming experience, aiming for a future where AI coding agents are accessible to all.