OpenAI on May 29 announced that Codex, its coding assistant, now supports Computer Use on Windows and remote control from iOS and Android devices and Macs, closing a platform gap two weeks after the feature debuted for Mac hosts. The move extends agentic capabilities—the ability for AI to see, click and type in desktop applications—to Windows, but with constraints that underscore the early stage of the technology.
Codex has been evolving from a code-file helper into an agent that takes direct action on a computer’s interface, a trend that competitors such as Anthropic are also pursuing. For enterprise developers and IT leaders, the Windows release means a broader pool of users can now tap into that workflow, but the rollout is far from universal.
The development comes as competitors including Anthropic pursue similar desktop agent capabilities, intensifying the race to turn coding assistants into doers. The new capability lets a user start a Codex task on a Windows host—editing files, running shell commands, navigating apps—and steer it remotely from a phone running ChatGPT or from a Mac. When a user sends a prompt from the mobile app, Codex executes it on the host, accessing local files and tools, and the user approves or tweaks the next step. The host must stay awake, online and signed in. Computer Use allows the AI to see the screen, click buttons and type into foreground windows, but it cannot operate in the background while someone is working in the same session; it takes over the active desktop and moves the pointer. Unlike the Mac version, which can run Computer Use with the screen locked, Windows requires the desktop to remain awake and unlocked throughout the task.
OpenAI emphasized that the feature is not a full remote desktop. It is unavailable at launch in the European Economic Area, the United Kingdom and Switzerland. Enterprise and Education customers can only access it after an OpenAI account representative enables the capability, and workspace administrators may need to approve remote connections. Additionally, a Windows machine cannot yet act as a controller for another computer; that capability remains limited to Macs.
The update builds on the March 4 release of the Codex app on Windows via the Microsoft Store and the May 14 launch of mobile Codex, which initially supported only Mac hosts. At that time, OpenAI said Windows support was “coming soon,” and the company reported that more than 4 million people use Codex each week, a figure tied to the overall platform rather than the new rollout.
Several details remain open. The specific Windows versions supported—whether the feature works on Windows 10 as well as Windows 11—have not been confirmed by primary OpenAI documents; technology publications have cited both. It is also unclear whether the release reached consumer-tier plans such as Plus or Pro on launch day, or if it is restricted to business and enterprise accounts for now. OpenAI has not said whether a background or locked-screen mode is planned for Windows, and the administrative security settings for enterprises are not yet described in public documentation. Additionally, how approvals for Computer Use actions appear on a small mobile screen—and what safeguards exist to prevent inadvertent authorisations—has not been outlined.
For companies tracking the agentic coding trend, bringing Windows into the Codex fold is a tangible move, but the foreground-only architecture and regional gating mean the tool remains closer to an assisted task runner than to an autonomous desktop agent. Enterprise buyers will need to weigh those boundaries against the convenience of extending AI-driven coding workflows to a broader operating system footprint. The update positions Codex as a broader tool for cross-platform development teams, but the restrictions signal that full autonomous operation is still several turns away.