Computer Use

Computer use is an agent capability where a model operates a computer the way a person does: it receives screenshots of the screen, reasons about what it sees, and emits mouse movements, clicks, and keystrokes to accomplish a task. Anthropic shipped the first frontier version in late 2024, and OpenAI's Operator followed; the pattern matters because it works on any software with a GUI — legacy ERPs, admin panels, sites without APIs — where classic integrations are impossible. Each step is a loop: capture screen, decide action, act, observe the result. That loop is slower and more expensive than calling an API, and misclicks compound, so production deployments sandbox the environment, restrict credentials, and keep a human approving irreversible steps like payments or deletions. For SaaS builders, computer use unlocks automation of the long tail of tools that will never expose an endpoint.

Related terms

More AI Agents terms