OpenAI just flipped a big switch for Windows users: Codex’s “computer use” is no longer a Mac-only trick. It can now actually see and operate apps on your Windows desktop, and you can even start and steer that work from the ChatGPT mobile app while your PC keeps chugging away at tasks in the background.

This isn’t just another coding assistant update – it’s your first real taste of AI that doesn’t just suggest what to do on your PC, it actually does it.

If you’ve missed Codex so far, think of it as OpenAI’s work-focused control center: an agentic environment where an AI can write and refactor code, run tools, juggle long-running jobs, and now, with “computer use,” directly manipulate your desktop like a highly skilled remote coworker. On macOS, that already meant Codex could read your screen, click buttons, fill forms, and move through GUIs instead of relying only on command-line tools or tightly integrated APIs. With the latest update, Windows finally joins that party: Codex can see your Windows desktop, interact with apps in the foreground, and carry out multi-step workflows inside real Windows software.

The basic idea behind computer use is simple but powerful: instead of saying “open this app, click here, type that” yourself, you describe the outcome you want, and Codex figures out the sequence of UI actions required. You might ask it to “pull the latest sales report from our internal dashboard, export it, and summarize it for the marketing team,” and Codex will navigate the browser window, operate the menus, and hand back the result. For tasks that live in awkward corners of enterprise software, half-baked web UIs, or internal tools with no public API, this approach is a big deal.

On Windows, there are some practical constraints: Codex needs the target app visible in the active desktop session, because it literally works from what it can see on screen. This aligns with how the feature works on macOS, where you grant screen recording and accessibility permissions so Codex can both view and interact with windows, menus, keyboard input, and clipboard state. In other words, this is closer to giving an assistant remote control of your PC than giving it a set of backend APIs. That sounds slightly scary, but it also makes the feature flexible in a way that traditional integrations often aren’t.

What really makes OpenAI’s announcement feel futuristic is the combination of Windows support and Codex inside the ChatGPT mobile app. Codex’s remote control started rolling out on mobile earlier this month, letting you see a live view of the machine Codex is working on, approve actions, and tweak prompts from your phone. Before, that was mostly a macOS story: your Mac could keep running builds, tests, and automations while you were out, with the ChatGPT app acting as your “mission control” for Codex. Now, that same pattern extends to Windows PCs: your desktop can be at home or under your desk at the office, while you manage Codex’s work from your pocket.

This mobile link-up matters for two reasons. First, it changes how tethered you are to your primary machine: you don’t need to drag a laptop everywhere to check on a deployment, re-run a test suite, or nudge a research task in a new direction. Second, it reframes Codex as something that runs on your machines, in your environment, using your tools and files, while you orchestrate it from wherever you happen to be. You’re not just “using an AI in the cloud” – you’re supervising an AI that’s literally driving your Windows desktop on your behalf.

From a day-to-day perspective, the obvious winners here are developers and technical teams, but the feature isn’t limited to hardcore coders. OpenAI frames computer use as the right choice whenever a task depends on a complex graphical interface that’s hard to verify via files or terminal output alone. That could be:

• Navigating a finicky Windows-only CRM that your sales team lives in
• Reproducing a weird bug that only appears in a specific GUI workflow
• Running QA across a mix of desktop and browser-based tools
• Automating setup steps in IDEs, test harnesses, or proprietary dev tools

On Windows, Codex can do this by “seeing, clicking, and typing in the foreground while it works,” which is exactly how humans interact with traditional desktop apps. When the internal system you rely on is locked behind legacy UI and no one wants to build an API for it, having an AI that operates the interface directly is a pragmatic hack.

It’s not hard to imagine less technical uses, either. A non-developer could ask Codex to log into vendor portals, pull invoices, reconcile entries in a finance tool, or routinely export data from an on-prem app that the IT team has never modernized. The combination of Windows support and mobile orchestration means a small business might eventually treat Codex as a sort of invisible office assistant living on a spare PC in the corner.

Of course, the idea of an AI agent driving your Windows desktop is a recipe for anxiety if it’s not wrapped in guardrails. Both the Codex docs and OpenAI’s broader tools guidance stress human-in-the-loop approvals, scoping tasks carefully, and imposing boundaries on what an agent is allowed to touch. OpenAI already recommends guardrails for tool use and side-effecting actions – checking arguments, gating sensitive operations, and pausing runs for human review before anything destructive happens. For computer use specifically, the docs are explicit: because Codex can affect app and system state outside your project workspace, you should use it for scoped tasks and review permission prompts before you continue.

That philosophy mirrors the company’s broader safety posture, where high-capability models – including the latest Codex variants – are treated as “high cybersecurity capability” and gated behind additional safeguards for sensitive work. Practically, this translates into flows where Codex proposes actions, but waits for you to approve them on desktop or mobile before it actually clicks the scary buttons. The new Windows support doesn’t change that dynamic; it just widens the surface area where those safety systems have to be done well.

There’s also the mundane but important question of what data Codex sees. By design, it’s working from screenshots and window contents, which can include sensitive information. Enterprises will likely end up pairing computer use with strict policy boundaries, separate Windows profiles, or dedicated VMs so that agents only see what they genuinely need. It’s not hard to picture organizations deploying Codex into controlled Windows environments the same way they sandbox RPA bots today.

Zoom out a little, and the Windows rollout feels like one more step toward a more agentic style of computing that has been creeping into view all year. First, we saw Codex as a kind of “coding command center” on macOS, with multiple AI agents sharing context and running long-lived dev tasks. Then came mobile access, which made that agent something you could supervise from anywhere. Now, with computer use working on Windows, the addressable surface expands from “Mac laptops and devboxes” to the largest desktop platform in the world.

Other players are pushing in similar directions – traditional RPA vendors, browser automation tools, and new wave “AI PC” pitches all talk about assistants that operate your apps for you. But Codex’s approach is slightly different: it’s less about recording rigid scripts and more about giving a generally capable model permission to act, then wrapping that in approvals, guardrails, and a unified app that keeps everything in one place. Add mobile “mission control” on top, and you start to get a picture of personal or team-specific AI that doesn’t just answer questions, but runs your actual workflows across devices.

Right now, OpenAI still describes these features as early or preview experiences, and there are regional limitations – computer use is available for macOS and Windows, but not in some regions like the European Economic Area, the UK, and Switzerland at launch. That’s a reminder that regulators are watching closely, especially when an AI sees what’s on your screen and can act on it. But it also signals something else: this is not a quirky side experiment, it’s part of OpenAI’s mainline product roadmap.

For Windows users, the headline is straightforward: Codex can now take real actions on your PC, not just give you advice about what you should do. You can kick off work from your phone, watch it unfold on your Windows desktop, and step in when something important needs a human decision. It’s still early, and it will almost certainly be messy at the edges – flaky UIs, odd layouts, and legacy tools are not going away. But if you’ve been waiting for a moment when AI starts feeling less like a chat window and more like a colleague who can actually drive, this Windows rollout is one of those inflection points.