Computer Use in Copilot Studio (Generally Available)

Copilot Studio agents can operate websites and Windows apps by clicking and typing, for systems that offer no way to connect directly.

You describe the task in plain language and the agent carries it out on a Windows machine you set up, driving a virtual mouse and keyboard
It reads the screen rather than following fixed coordinates, so it keeps working when buttons or layouts move
You pick the model behind it: OpenAI's Computer-Using Agent and Claude Sonnet 4.5 are generally available, with Claude Sonnet 4.6 and Opus 4.6 experimental
Each step costs 5 Copilot Credits, or 15 on a premium model, so a four-step form fill runs about 20 credits
You can restrict it to an allow list of sites and apps, require HTTPS, and nominate someone to email if it spots instructions that look like an attempt to hijack it
Needs generative orchestration turned on, and it suits autonomous agents running in the background more than conversational ones

Automating a system with no integration available, like an aging internal tool or a vendor portal

The work is repetitive screen work such as data entry, invoice processing, or pulling figures out of a dashboard

Ops and marketing teams stuck doing manual work in systems IT cannot connect

Join the newsletter to get ahead with AI.

I share strategies, new modules, free guides, and tools to get you better at using AI in your work.