Oct 2024
Anthropic Computer Use — Claude Controls Desktops Autonomously
Anthropic released a Computer Use capability for Claude 3.5 Sonnet in public beta, enabling the model to view screenshots and operate computers by moving a cursor, clicking interface elements, and typing text — without requiring custom API integrations for each application. This represented a significant step toward general-purpose computer agents that could automate any software task a human can perform on a screen. Anthropic flagged the capability as experimental and susceptible to prompt injection.
Why it mattersScreen-level computer control means AI agents can automate any GUI-based workflow without API access, unlocking automation for legacy enterprise software.
Try itUse Anthropic's computer use API to automate a repetitive GUI task in your stack — start with a well-defined, low-stakes workflow like form filling or data export to evaluate reliability.