In a post dated 26 May 2026, Microsoft said computer-using agents in Copilot Studio are generally available, with embedding in multi-step workflows moving into preview. These agents interact with websites and desktop applications through the user interface.
From the announcement
- They are aimed at processes where there is no API.
- New enterprise capabilities include more secure credential management, choice of models for different scenarios and automations that adapt to changing interfaces.
- A customer example, a service-order agent for a relocation company, is described as live; results are described qualitatively with no figures.
Governance questions
- Whose credentials does the agent use, and how are they stored and rotated?
- What can it do on screen, and which actions require approval?
- How are its actions logged so that they can be reviewed?
- What happens when the interface changes and the agent clicks the wrong thing?
Screen-driven agents are a bridge for legacy systems, not a substitute for proper integration where an API exists.