TL;DR
- Windows Update: OpenAI Codex has added computer use and phone-based control for Windows users.
- Desktop Limit: Codex runs on the active desktop, so users cannot keep the same Windows session available during automation.
- Phone Oversight: ChatGPT on a phone can review output, approve actions, and send follow-up instructions to a connected PC.
- Workflow Context: The Windows release combines earlier Mac computer use and mobile supervision into one desktop-plus-phone workflow.
An update for OpenAI Codex for Windows update has added computer use and phone-based task control. Users can new steer desktop tasks from a phone while the work continues on a PC.
OpenAI is combining the ChatGPT mobile app rollout for Codex with the GUI automation already available on the macOS app. Together, those pieces turn Codex into a desktop-plus-phone workflow rather than a coding assistant confined to chat or an editor.
Windows reached this point after the Codex app Windows release of March. OpenAI has added a stricter Windows sandbox that tightens how the app operates locally. Developers now get a feature set built around agent safety, permission boundaries, and session handoff instead of raw desktop access.
What the Windows rollout adds
Codex can operate Windows desktop apps by reading the screen, clicking interface elements, and typing through a task flow. Codex can use that capability to test interfaces, step through bugs, and review work where the project context already lives on the machine.
Windows users, this one’s for you.
Computer use now works on Windows, so Codex can take action on your Windows computer.
And with Windows support for Codex in the ChatGPT mobile app, you can start, review, and steer tasks on the go while work continues on your Windows machine.… pic.twitter.com/OPIxOcP4Nl
— OpenAI (@OpenAI) May 29, 2026
Deliberate runs such as GUI testing, installer checks, or bug reproduction fit that model better than quiet background jobs. Foreground operation remains the main practical limit. Codex runs on the active desktop, so users cannot keep using the same session normally while it is controlling another app.
That limitation means the Windows machine still has to be treated as the task surface, not as a second screen where normal work continues in parallel. Before the newer sandbox work, the platform left users choosing between command approvals and full access. The improved guardrails now make wider desktop automation possible, but they do not remove the need to hand the active session over to the agent while the task is running.
Phone-based supervision is the other major part of the release. Users can connect a PC to Codex from ChatGPT on iPhone or Android, then review approvals, diffs, test results, screenshots, and terminal output without returning to the desk. They can also send follow-up instructions or start a new thread against the connected host from the phone.
Because the work happens on the Windows (or Mac) host, the phone becomes a review surface rather than a remote execution environment. Approvals and screenshots stay in front of the user while Codex performs the task on the PC. For a developer away from the desk, that means a build or UI check can continue without losing the approval point.
Users can mention @Computer or an @AppName in the prompt after installing the Computer Use plugin in settings. ChatGPT on mobile can check in on tasks or start new tasks against the connected host.
That remote workflow depends on a connected Windows/Mac machine staying awake, online, and signed into the same account long enough for the job to complete. If the machine sleeps, signs out, or drops the account session, the handoff point disappears. For deliberate task runs, that setup works better than unattended background automation.
Competition around coding assistants remains intense. Organizations already standardized on GitHub Copilot more naturally, while Claude Code remained the stronger fit for large multi-file planning work in the same comparison.
