Computer and browser use

Agents that operate GUIs through screenshots, clicks and typing, for systems without APIs.

Changes fast: check the latest updates too. Reviewed 17 Oct 2026.

Key ideas

• Prefer APIs and structured page reading (accessibility tree, DOM) over pixel clicking when available.
• Run in isolated VMs or browsers with no saved credentials beyond what the task needs.
• Expect slower, more expensive runs; use for long-tail tasks, not high-volume ones.

Common mistakes

• Exposing a computer-use agent to untrusted websites with access to sensitive accounts.