OpenAI Codex just got a new trick: it can now open a browser, click around your front-end code like a human tester, and fix bugs it finds along the way. After developers prompt Codex to generate front-end code, the coding agent fires up a built-in browser, interacts with the rendered UI, and uses visual inspection plus network and console logs to identify and patch issues — all without you needing to manually switch tabs to verify the result.
https://twitter.com/JamesZmSun/status/2047522852854026378
The feature runs through Codex's new Browser plugin, and users can manage allowed and blocked sites in settings. According to OpenAI Codex engineer James Sun, this update brings Codex "closer to a fully autonomous coding agent that can deliver high-quality, tested code changes." This follows other recent Codex updates, like its screen-watching memory feature.