OpenAI announced Codex: a cloud AI agent for software tasks capable of handling multiple assignments in parallel. Codex can write features, answer questions about your code, fix bugs, prepare pull requests, and run tests in an isolated environment preloaded with your repository.
Codex is built on codex-1 (OpenAI version o3, tailored for programming) and was trained on real development tasks to closely mimic the style and preferences of programmers. Special attention was given to ensure patches are clean and ready for review and integration.
Codex is already available for ChatGPT Pro, Team, and Enterprise, with support for Plus and Edu coming soon. Tasks can be launched directly from ChatGPT via the sidebar, allowing you to monitor progress, receive logs and test results, review changes, or integrate them immediately.
🔗 Source: https://openai.com/index/introducing-codex/
▶️ Demo: https://www.youtube.com/watch?v=hhdpnbfH6NU
P.S. By the way, it now makes sense why o3 and o4 were constantly outputting diffs instead of code. It was a move towards agents that will apply these diffs.
