Microsoft Tests New Copilot Feature That Lets AI Use Apps Like a Human Part of an early access research preview, the feature is aimed at automating repetitive tasks that usually require human input. This could include entering data into older desktop applications, collecting information from websites, or performing basic tasks like invoice processing

Microsoft has launched an experimental feature in Copilot Studio called Computer Use, which enables AI agents to interact with websites and desktop applications through the user interface—much like a human would. Rather than relying on APIs or backend integration, the AI performs tasks by clicking buttons, filling out forms, and navigating drop-down menus directly on the screen.

Part of an early access research preview, the feature is aimed at automating repetitive tasks that usually require human input. This could include entering data into older desktop applications, collecting information from websites, or performing basic tasks like invoice processing.

What sets this feature apart from traditional robotic process automation (RPA) tools is its ability to adapt when something changes on the screen, such as a button moving or a webpage layout shifting. These kinds of changes often disrupt older automation systems, but Microsoft says Copilot agents can continue functioning without manual intervention.

The system runs through Microsoft's cloud services, meaning users don't need to manage their own servers to run these AI agents. Microsoft has also stated that enterprise data accessed by Copilot isn't used to train its large language models, addressing some common concerns about data privacy.

The tool is being positioned for tasks like automating data entry, pulling information for market research, or processing digital invoices—areas where businesses often rely on manual labor due to lack of integration options. As it stands, the feature is still in its testing phase and its practical reliability in diverse environments remains to be seen.