Microsoft Unveils New AI Feature in Copilot Studio for Task Automation

Microsoft on Friday debuted a new research-preview capability in Copilot Studio that lets AI agents operate websites and desktop applications as a person would, without relying on an API.

Dubbed "computer use," the feature allows users to describe a task in ordinary language and watch the AI agent perform it by clicking buttons, entering text and navigating menus. Organizations can test the sequence and fine-tune the steps before deploying agents across browsers such as Microsoft Edge, Google Chrome and Mozilla Firefox, as well as native Windows applications.

"If a human can complete the workflow, the agent can too," said Charles Lamanna, corporate vice president for Microsoft’s Business and Industry Copilot. He added that computer use extends automation reach to interfaces that lack formal programming hooks.

Built for real-world operations, the feature supports activities such as bulk data entry, market analysis and invoice handling. Companies can feed information from multiple sources into centralized systems, reducing manual errors and speeding up routine processes.

Unlike tools that stall when an element shifts or a CAPTCHA appears, Microsoft’s solution uses internal reasoning to adapt to screen changes and continue the task. Every action, along with screenshots and decision logs, is recorded in an activity history for auditing and review.

Microsoft confirmed that all data processed stays within Microsoft Cloud boundaries and will not train its Frontier AI models. Running on Microsoft-hosted servers, the computer use feature removes the need for enterprises to deploy and maintain their own infrastructure.

Early access to computer use is open now to select Copilot Studio customers. A broader rollout is planned for later this year, as part of Microsoft’s push to expand AI-driven automation in business environments.