AI AgentsInfo

Glossary

Computer use

A tool loop in which a model looks at a screen and asks for a mouse or keyboard action, and the app runs that action.

Computer use is a tool loop. The model looks at a screen and asks for a click, a keystroke, or a scroll. The app around the model takes the screenshot, runs the action, and returns the new screen. That app is part of the harness. The working pair is an agent.

Anthropic's computer-use tool is the API form: one toolset entry, and your application runs each call in an environment you control. Claude's product article describes the same idea in Claude Cowork and Claude Code, on a Mac or Windows desktop. A check before a risky click is a guardrail. How far the loop runs before a person confirms is autonomy.

A narrower tool that stays on a web page is browser use. The longer guide is Can an AI agent use your computer?.

Related terms