Claude learns to operate a computer
Anthropic releases "computer use" in beta: Claude can look at a screen, move the cursor, click and type to complete tasks.
A historical account of this event. Usage notes are YLEM editorial analysis; check the provider for current features and prices.
What changed
The public API beta lets Claude inspect screenshots and issue cursor, click and typing actions. Anthropic described it as experimental and error-prone, including difficulty with familiar interface operations.
Sources: [1] Anthropic · Computer use beta
What this means in practice
Start with a reversible workflow, such as collecting information into a draft. Give the agent a clear stopping point and check that the resulting file or page actually contains the requested information.
How to read the result
Operating an interface is different from understanding the consequences of an action. Use human review before consequential submissions, and keep a record of the steps completed.