GitHub adds computer-use preview to Copilot CLI and desktop app

Builders can enable computer use via /computer in Copilot CLI or the desktop app settings and test it on tasks that currently require manual GUI steps. An illustrative scenario would be automating data entry from a legacy internal application into a modern dashboard: the builder would grant Accessibility and Screen Recording permissions, run the agent on sample records, and measure success rate against manual time. Decision criteria include whether the interface remains stable enough for the workflow and whether saved approvals reduce friction without violating enterprise policy. Unknowns are exact failure modes on specific applications and long-term reliability under changing window states.

Operators should review managed-settings.json and enterprise policy before enabling computer use, especially for agents that may encounter sensitive on-screen data. An illustrative scenario would be trialing the feature on a non-customer-facing reporting workflow: the operator would configure approval prompts, run a small set of tasks, and log any instances where the agent accessed unintended windows or required manual intervention. Decision criteria focus on whether the productivity gain justifies the added monitoring overhead and whether data visible on screen stays within approved boundaries. Unknowns include the scope of logged interactions and any retention of screenshots.