Microsoft agents can now click and type in apps like a human
Reviewed by Helen Jones
Microsoft updated its Copilot Studio documentation for computer use, a tool that lets AI agents automate websites and desktop apps by clicking buttons, choosing menus, and typing into fields. The agent works from natural language instructions and uses a virtual mouse and keyboard on a computer you set up. Usage is billed as an Agent action feature at 5 Copilot Credits per step, or 15 credits when premium models are used. Supported models at the standard 5-credit tier include OpenAI Computer-Using Agent and Anthropic Claude Sonnet 4.5. The tool can hand off to Power Automate desktop flows for more precise execution.
Until now, most citizen makers stuck to cloud flows with SharePoint, Outlook, Teams, and Forms, or used desktop flows when a process lived only in a screen. UI automation was either brittle recorded steps or something left to specialists, and quiet failures already made people distrust anything that ran unattended. Computer use changes the option set. Agents can adapt when a button moves, which is the classic desktop-flow failure mode, but every click and keystroke now burns credits and still needs agent setup, ownership, and a clear failure path. For people building approvals and request routing, this is a hybrid layer, not a free upgrade to the flows they already struggle to monitor.
Analysis
Treat this as a trend to watch, not a reason to rebuild working cloud flows. Pick one sticky screen-only step that already frustrates a desktop flow, estimate steps times 5 credits, and only then test a small agent handoff with a human review path and a clear owner.
Source note
Pulse published by Collab365 Spaces, reviewed by Helen Jones on . Cite as "Microsoft agents can now click and type in apps like a human", Collab365 Spaces. 2 sources referenced.