• Microsoft’s Copilot Studio computer-use agents hit general availability on May 13, making it the first major platform to ship the capability at production scale 
• Anthropic and Google are still in beta and preview, giving Microsoft a window that will not stay open forever

There is a problem most enterprise IT teams know well but rarely talk about. The software running core business operations — vendor portals, ageing internal tools, line-of-business applications built a decade ago, has no API. No integration shortcut. Someone has to log in, click through screens, and handle it manually.

Microsoft Copilot Studio’s computer-use agents are built to change that. Now generally available across all commercial geographies in Microsoft Power Platform, the capability lets AI navigate software the same way a person does.

The announcement, authored by Copilot Studio principal product manager Mustapha Lazrek, confirmed what the preview had pointed toward: agents can now interact with any application that has a graphical interface, clicking buttons, selecting menus, and typing into fields.

Charles Lamanna, president of business applications and agents at Microsoft, put it simply: “If a person can use the app, the agent can too.” For enterprise IT, that carries real weight. Traditional automation has always depended on either well-maintained RPA scripts or APIs that the legacy system was never designed to expose. 

When the interface changes, the script breaks. When the vendor offers no integration path, the workflow stays manual. Computer use in Copilot Studio works around both by operating through the visual layer itself, using reasoning to navigate whatever happens to be on screen.

What actually shipped

The GA release is a meaningful step up from the September 2025 preview. Four things were added: model choice across OpenAI and Anthropic (with Claude Sonnet 4.5 and OpenAI CUA confirmed as the production-supported models), Azure Key Vault credential storage, Microsoft Purview audit log propagation, and Windows 365 Cloud PC pool support for isolated execution environments.

The audit trail piece is arguably what makes the rest usable at enterprise scale. Organisations deploying agents into live business systems need visibility into what the agent did and why. Copilot Studio now logs sessions to Purview and Dataverse, with session replay carried forward from a January 2026 preview update. Human-in-the-loop checkpoints can be configured for steps that require approval or fall below a confidence threshold. DLP policies and environment isolation extend from the existing Power Platform governance stack.

That combination, an auditable, governable agent that can operate inside systems with no API, is genuinely new at the production scale.

Where things stand competitively

Microsoft is currently the only platform with contractual general availability for computer-use agents. As of this week, Anthropic’s Claude Computer Use remains in a paid-plan beta, and Google’s Gemini Computer Use is in public preview. Neither offers enterprise buyers the combination of a production SLA, compliance-grade logging and geo-broad availability that Copilot Studio now ships with.

That lead is temporary. Both Anthropic and Google have active development programmes in this space, and the gap will close. But for organisations running procurement decisions or planning automation roadmaps right now, Microsoft holds a concrete head start.

The limits worth knowing

The capability is Windows-centric in its current form. Sovereign cloud environments serving US government customers are excluded from the rollout. Pricing runs on Copilot Credits at five credits per step, which works out to roughly US$0.04 per step on the standard prepaid pack. A four-step workflow costs around US$0.16. That is fine for targeted use cases, but high-volume deployments need proper cost modelling before they scale.

The workflows that make the strongest case for computer-use agents in Copilot Studio are the ones with high manual frequency and no API path: invoice processing through vendor portals, updating records in legacy systems, pulling data from internal tools that pre-date modern integration standards. 

These have sat outside the reach of enterprise automation for years. That is now a narrower category than it was two weeks ago.

Want to experience the full spectrum of enterprise technology innovation? Join TechEx in Amsterdam, California, and London. Covering AI, Big Data, Cyber Security, IoT, Digital Transformation, Intelligent Automation, Edge Computing, and Data Centres, TechEx brings together global leaders to share real-world use cases and in-depth insights. Click here for more information.

TechHQ is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.