Archive · 15 Dec 2025
OpenAI's Codex AI Writes Majority of Its Own Code
OpenAI has an AI called Codex that helps write and improve its own code.
Most of OpenAI's developers use Codex to write new features, fix bugs, and review code.
Codex can even plan and write code for apps like the Sora Android app.
It works like a junior developer, but humans still check all the code.
OpenAI says Codex makes developers faster and catches bugs before they cause problems.
Other companies also have similar AI tools, but OpenAI's latest version is doing better on tests.
Codex is meant to help developers, not replace them.
OpenAI's AI coding agent, Codex, writes the majority of its own code under developer supervision.
Over 92% of OpenAI's technical staff use Codex for tasks like writing new features and fixing bugs.
Codex reviews every internal pull request and helps engineers ship 70% more merged pull requests.
Codex wrote 80% of the Agent Builder tool in less than six weeks.
OpenAI faces competition from other AI coding tools like Anthropic’s Claude Code and GitHub Copilot.
- Who
- OpenAI and its AI coding agent, Codex
- What
- Codex writes the majority of its own code under developer supervision
- When
- Disclosed in 2025
- Where
- OpenAI's development process
- Why
- To improve efficiency and assist developers in coding tasks
- AI Agent
- Codex
- Developer Usage
- 92% of OpenAI's technical staff
- Pull Requests
- 70% more merged pull requests with Codex
- Bug Detection
- Catches bugs before production
- Agent Builder Tool
- 80% written by Codex in 6 weeks
- Competitors
- Anthropic’s Claude Code, GitHub Copilot, Google’s Gemini CLI, Cursor
- Benchmark Score
- GPT-5.2 scored 55.6% on SWE-Bench Pro
- Human Oversight
- Developers review all code