Experimental Copilot Actions lets testers control a virtual browser

Microsoft’s Copilot Actions feature has started functioning in what appears to be an early labs release. This feature allows users to trigger a virtual environment, essentially a cloud-based machine with a browser, where Copilot can directly interact with web pages. The interface opens as a split-screen: a desktop-style window in the center with Copilot chat on the right side. What sets it apart is a unique “Take Over” button that enables the user to manually control the mouse, effectively letting them interact with the remote desktop as if it were their own system.

This is similar in nature to ChatGPT’s Operator feature but may end up being more accessible. While Operator is exclusive to ChatGPT Pro users, Copilot Actions could be made available for Copilot Pro subscribers, which is notably less expensive. Although early assumptions pointed to Windows-only support, the current browser-based implementation indicates a broader compatibility regardless of operating system. That opens the door to wider adoption, though performance is still a major constraint. The startup and action execution phases remain slow, suggesting Microsoft is still refining its backend infrastructure. The experimental nature of this rollout makes it unlikely to see general availability soon, but it hints at a future where task delegation to AI within live environments becomes commonplace for Copilot users.