OpenAI has rolled out a major update to its Codex coding agent, a move seen as laying crucial groundwork for its anticipated desktop super app, which aims to unify ChatGPT, Codex, and the Atlas web browser. While the super app itself is not yet available, this Codex update significantly expands its capabilities, offering developers a glimpse into OpenAI’s future vision.
Thibault Sottiaux, head of Codex, stated during a press briefing, “We’re building the super app out in the open. This release is about developers. In the future, we will broaden it up to a wider audience.” The latest version of Codex now provides developers with multi-purpose AI agents that can operate across a “larger surface area” and exhibit greater proactivity.
Key new capabilities include:
- Computer Use: Codex agents can now interact with other applications on your PC. When prompting, users can specify a program or allow the AI to determine the best application for the task. OpenAI highlights its “secret sauce” that enables agents to run apps without bogging down the entire system, facilitating tandem work.
- 111 New Plugins: The update introduces 111 new plugins that combine skills, app integrations, and model context protocol server connections. These significantly enhance Codex’s ability to gather context and utilize tools critical for developers' workflows.
- Built-in Browser with Commenting System: Codex now features an integrated web browser with a commenting system. This allows users to prompt Codex to make specific tweaks to webpages or web applications being built, such as adjusting graph margins to prevent axis cutoff.
- Built-in Image Generation: Codex leverages gpt-image-1.5 to create product concepts, mockups, frontend designs, and even assets for simple games. It can also use screenshots to verify its progress against user requests.
- Memory Features (Preview): OpenAI is previewing two memory-related features. The first allows Codex to recall context from previous tasks to inform future prompts, aiming for faster and higher-quality request completion. The second enables the app to proactively suggest actions based on gathered context, such as prompting a user to respond to a coworker's comment on a Google Doc draft.
The updated Codex is now rolling out to desktop app users logged in with their ChatGPT accounts.