Microsoft ships Copilot Studio GA

Microsoft has moved Copilot Studio’s “computer use” feature into general availability, giving enterprise customers a way to build agents that click through websites and desktop applications rather than relying only on APIs. The release, announced May 13, extends Copilot Studio into the long tail of internal tools, vendor portals and line-of-business software that still require human navigation. Microsoft said the feature is available across all commercial geographies in Power Platform, with data residency and compliance controls tied to the customer’s tenant. The company framed the release as part of a broader push to make agents more governable inside enterprise environments. ### What exactly became generally available? Microsoft said on May 13 that “computer use” in Copilot Studio is now generally available, allowing makers to add a tool that can operate “inside any application a person can use.” In Microsoft’s product description, the agent works on a Windows computer, reads the screen, and carries out actions such as clicking buttons, choosing menus and typing into fields. (techcommunity.microsoft.com) Microsoft Learn describes the feature as a way to automate web and desktop apps through natural-language instructions. The documentation says the agent can complete tasks even when there is no direct API connection to the target system, including data entry, invoice processing and extraction work. ### Why does this matter for older enterprise software? (techcommunity.microsoft.com) Microsoft said the gap it is targeting is the set of workflows “buried in vendor portals, internal web apps, and proprietary line-of-business systems” that have remained outside modern automation. That includes systems where companies either had to build and maintain robotic process automation scripts or wait for APIs that never arrived. (learn.microsoft.com) The practical change is that a Copilot Studio agent can now be pointed at UI-driven work such as filling forms, navigating admin consoles or completing follow-up tasks after a meeting, so long as a person could perform the same sequence on screen. Microsoft’s FAQ says the tool interprets screen pixels, reasons over the current state, and then performs mouse and keyboard actions in an iterative loop. (techcommunity.microsoft.com) ### How does Microsoft say the agent actually works? Microsoft Learn says computer use is powered by a Computer-Using Agent model that combines vision with reasoning to interact with graphical user interfaces. The system captures screenshots, evaluates the current screen state and prior actions, and then decides whether to click, type or scroll. (learn.microsoft.com) Microsoft also says customers can choose the model used to execute the tool. The current supported options listed in Learn are OpenAI’s Computer-Using Agent and Anthropic’s Claude Sonnet 4.5, with Anthropic use dependent on an administrator enabling external models. ### What controls did Microsoft add for enterprise customers? Microsoft said the GA release includes allow lists for websites or desktop applications, built-in credential handling with Azure Key Vault, and native Power Platform controls such as data loss prevention policies, environment isolation and audit trails. (learn.microsoft.com) The company also said low-confidence steps and exception cases can be routed through human approval checkpoints. Microsoft’s documentation says run history shows what the agent saw, clicked and why, while logs are propagated to Microsoft Purview and Dataverse for audit and administrative review. The company’s FAQ adds a warning that ambiguous instructions or unexpected on-screen content can lead to unintended actions affecting devices, data or connected accounts. (techcommunity.microsoft.com) ### Is every part of the stack already production-ready? Microsoft’s run-time options are still mixed. Learn documentation says computer use itself is generally available, but the hosted browser and Cloud PC pool deployment options are still in preview and powered by Windows 365 for Agents. Microsoft says the hosted browser is intended for quick starts and experimentation, is not recommended for production use, and may be throttled. (techcommunity.microsoft.com) Microsoft said customers can also run computer use on their own Windows machines. The next step for buyers is likely to be choosing where these agents run and how much approval and audit coverage they want before using them on internal systems. (learn.microsoft.com)