OpenAI has made a major update to the Codex AI coding assistant today. This upgrade focuses on three core areas: Mac desktop application control, parallel operation of multiple agents, and the ability to handle tasks over longer time frames.
The most notable highlight of this update is that Codex now has the ability to directly operate Mac desktop applications. It has its own cursor, can independently view the screen, click on interface elements, and input text, all without any manual intervention. Developers can use this for application testing and front-end interface iteration. Multiple agents can also run in parallel, operating independently without disturbing the user's normal workflow.
In terms of task continuity, Codex has added a memory function that can save user preferences, common workflows, and tech stack information. After a task is paused, it can seamlessly resume progress through the existing conversation thread, even autonomously schedule work plans for days or weeks ahead, and proactively suggest ways to move forward based on project context and connected plugins.
The browser experience has also been enhanced. The in-app browser allows users to add annotations directly on pages, providing more accurate execution instructions for the agent. OpenAI revealed that full browser control capabilities are coming soon, at which point Codex will be able to independently open websites, perform operational processes, and take screenshots to verify results.
