Computer use is a newer AI capability where the model controls a computer the way a human would — moving the mouse, clicking, typing into apps, navigating websites. Anthropic introduced this in 2024-2025. Here is what it actually means and when to care.
Computer use is an AI capability where the model can see the screen and take actions (mouse, keyboard, navigation) directly in the operating system or browser.
In practice: instead of you telling the AI what to do step-by-step, the AI can complete entire workflows by operating apps the same way you would.
Simple, well-defined web tasks work. Fill out a form, navigate a website, extract data from a page.
Tasks in apps with consistent UI work. Microsoft 365, Google Workspace, common web apps.
Complex multi-app workflows are still flaky. Reliability drops when the agent has to switch between apps or handle unexpected UI states.
Anything time-sensitive or destructive needs verification. Computer use can buy things, send emails, etc. Always include safety constraints.
Data extraction from systems without APIs. Old enterprise software you cannot integrate via API.
Automated testing. AI driving your UI to test workflows.
Cross-app workflows that resist automation. When integration costs would exceed the savings.
Most B2B teams do not need computer use directly. The capabilities are evolving fast; check back quarterly.