OpenAI Codex Major Update: Can now directly manipulate the Mac desktop, and has the ability to automatically wake up across days to continue working

According to MotionBeat monitoring, OpenAI has released a major update to Codex, with over 3 million developers now using the product weekly. The key change in this update is that Codex is no longer limited to coding and has started to participate in the entire software development process.

The most significant feature leap is "Background Computer Control": Codex can now directly view the screen, click the mouse, and type on a Mac, operate any application, and multiple agents can work in parallel, each with its own cursor, without disrupting what the user is currently doing. For developers, this means that when iterating on the frontend interface, testing applications, or operating tools without APIs, they no longer have to manually describe what to do with screenshots; the agents take care of it themselves.

Within the application, a new browser has been added, allowing users to annotate comments on the page and give instructions to the agents directly, suitable for rapid iteration in frontend and game development. In terms of image generation, gpt-image-1.5 has been integrated, allowing the generation of product concept images, UI design drafts, and game assets in the same workflow. Over 90 new plugins have been added, covering development toolchains such as Atlassian Rovo (JIRA management), CircleCI, CodeRabbit, GitLab Issues, Microsoft Suite, and Neon by Databricks.

The expansion of automation capabilities is worth highlighting. Codex can now reuse existing conversation threads to retain context, schedule future tasks on its own, and automatically resume execution, with time spans of days or even weeks. A preview version of the memory function has also been introduced, which can remember user preferences, corrections, and information accumulated from historical operations, allowing subsequent tasks to be completed faster. Based on project context, connected plugins, and memory, Codex will actively suggest where users should start their day's work or continue with a previous project, such as identifying pending comments in Google Docs, pulling relevant context from Slack, Notion, and code repositories, and generating a prioritized action list.

In terms of developer workflow, new features include GitHub PR review comment handling, multi-terminal tab support, remotely connecting to a devbox via SSH (alpha stage), and sidebar file previews (supporting PDFs, spreadsheets, slides, and documents).

These updates are gradually being rolled out to Codex desktop users who are already logged in to ChatGPT. The memory and proactive suggestion features will be later extended to the Enterprise Edition, Education Edition, and EU/UK users, while computer control currently only supports macOS.