OpenAI Brings Codex Computer Use to Windows With Mobile Steering

OpenAI shipped a major Codex update on May 29 that brings Computer Use to Windows for the first time and lets you steer long-running tasks from your phone. The release closes a months-long gap between Mac and Windows users and pushes Codex closer to the always-on agent OpenAI has been promising.

The headline change: Codex can now see what's on your Windows desktop, move the pointer, click buttons, and type into native applications while it works. Before this, Computer Use was a Mac-only capability that turned Codex from a coding chatbot into something that could actually operate the machine it ran on.

What actually changed

Codex app version 26.527 is the build that flips Computer Use on for Windows. Once enabled, eligible users can ask Codex to do things like test a desktop app, exercise a GUI workflow, or wire up a third-party tool that doesn't expose a clean API.

There's a catch: on Windows, Computer Use runs on your active desktop. It can't operate in the background while you keep using the machine for something else. When Codex is driving, expect your cursor to move on its own, windows to come to the foreground, and your screen to behave like someone else is at the keyboard.

The update also adds remote continuation from mobile or a Mac. You can kick off a Codex task on your Windows PC and then leave the room — checking progress, approving steps, or redirecting the agent from the ChatGPT iOS or Android app, or from Codex on a Mac. That continuity is what makes the Windows desktop usable as a background worker even with the foreground constraint.

OpenAI also rolled in faster browsing performance, stability fixes, and a new Codex Profiles surface that shows usage, token activity, and credit consumption per profile. For teams managing how much Codex is spending across projects, that's the first real per-project visibility OpenAI has shipped.

Why this matters

Computer Use is the feature that separates an agent that talks about doing work from one that actually does it. Most enterprise workflows still live in desktop apps — IDEs, design tools, SAP clients, internal Win32 software — that no SaaS API can reach. An agent that can use those apps the way a person does is qualitatively different from one that can only call HTTP endpoints.

Windows is also where most enterprise desktops actually run. Mac-only Computer Use was useful for developers and demos, but the customers buying ChatGPT Enterprise are running Windows fleets. Bringing Computer Use to Windows is the move that lets OpenAI sell Codex into the operations workflows it's been pitching at Codex Profiles.

The mobile steering piece is the other half of the story. You can now start a Windows-resident Codex task from your desk, walk away, and approve or redirect it from your phone in a meeting. That's the workflow OpenAI has been describing as "background work" — long-running agentic tasks that don't need you watching the screen, just occasionally weighing in.

Industry impact

The release lands directly in the path of Microsoft Build, which opens June 2 with a Windows Agent Framework keynote from Satya Nadella. Microsoft has been positioning Windows itself as the platform for agents, with new APIs that let agents plug into the shell, task scheduler, and security model. OpenAI just shipped a Windows agent that doesn't need any of that — it drives the OS through the same input layer a human uses.

The competitive dynamic is awkward. OpenAI is Microsoft's anchor model partner, but Codex on Windows competes directly with GitHub Copilot agent mode and the Windows Agent Store features Build is about to announce. The two companies are now shipping overlapping agent products into the same surface.

Anthropic's Computer Use, launched in late 2024 and refined through Claude 4 and 4.5, has been the reference implementation other models chase. Codex's Windows release is OpenAI catching up where it had been visibly behind — and showing it intends to lead on the developer surface in particular.

Expert perspectives

Independent reviewers have been blunt that Computer Use on the active desktop is operationally limiting. Windows Forum noted that Codex "takes over the foreground" while it runs, which makes it impractical to use a single workstation for both human and agent work at once. The workaround most early adopters are discussing is dedicated Windows machines or VMs running Codex as a remote worker.

What's next

Computer Use on Windows is not yet available in the European Economic Area, the UK, or Switzerland. OpenAI has not given a timeline for European rollout, which is typical of Computer Use features given the EU AI Act's transparency and oversight requirements for autonomous systems.

The next watch is whether OpenAI announces a Codex on-premises offering. Snowflake's $200 million OpenAI partnership embedded GPT-5 family models inside Cortex AI in February, and Dell's Codex enterprise on-premises pilot launched May 18. If Codex Computer Use can run inside an enterprise's own Windows VDI fleet, it becomes a serious alternative to the RPA platforms it's quietly replacing.

The other thing to watch is the GitHub Copilot agent mode reveal at Build. Microsoft is expected to show off Copilot driving VS Code autonomously, with sub-agents for testing, documentation, and review. Whether that pitch survives a side-by-side comparison with Codex doing the same work — and steering Windows apps it doesn't have an extension for — is the question the next two weeks will answer.

The bottom line

Codex Computer Use on Windows is the unlock that takes OpenAI's agent product from "interesting for developers on Macs" to "deployable on the operating system enterprises actually run." Combined with mobile steering and per-profile usage tracking, it's the most complete autonomous coding agent OpenAI has shipped — and it lands three days before Microsoft tries to claim the Windows agent narrative for itself.