OpenAI said on Thursday it has rolled out a major update to its coding agent, Codex. The update allows Codex to automate a broad range of development tasks, including writing code, running tests, responding to GitHub reviews and integrating collaboration tools.

The key addition is a "background Computer Use" feature. While Codex directly operates macOS apps in the background, users can continue other tasks without interruption, and multiple agents run in parallel at the same time.

It can also handle repetitive tasks such as modifying code and running tests. OpenAI said it can resume interrupted work and continue long-running tasks that span several days.

Support for the macOS desktop app has also been expanded. If users instruct UI changes in an in-app browser, the agent carries out code changes based on those instructions, widening its use beyond the existing terminal-focused environment.

Collaboration features have also been strengthened. Based on the context of collaboration tools such as Slack, Gmail and Notion, Codex can track work or suggest follow-up tasks. It can also use the image generation model gpt-image-1.5 to handle UI design, mockups and game production within a single flow alongside code.

OpenAI is also providing an agent software development kit that lets users build and operate agent-based AI directly. It is designed to allow users to control agent behavior and run it reliably through a sandbox-based execution environment.

OpenAI said, "Codex is rapidly expanding its scope of use, starting with writing code to understanding systems, grasping context, debugging and managing long-running tasks." It added, "The goal is to reduce the gap between what users imagine and what can actually be implemented."