Anthropic’s Claude Code and Cowork can control your computer

Anthropic has released a significant update to its Claude AI, enabling its specialized Code and Cowork agents to autonomously perform tasks on a user's computer. Currently available as a research preview for Claude Pro and Max subscribers, this feature is limited to macOS devices and requires both the Claude desktop app and a paired mobile app to function. The system allows Claude to open files, operate web browsers and applications, and run developer tools directly, with Anthropic emphasizing that it requires no prior setup from the user and can even operate when the user is away from their machine.

This capability represents an evolution of the autonomous functions introduced with the Claude 3.5 Sonnet model in 2024, now specifically applied to programming and workplace assistance. The AI operates by first attempting to use direct connectors to services like Slack and Google Workspace. If a connector isn't available, it defaults to controlling the computer's browser, mouse, keyboard, and display directly. A critical safety feature is that Claude will always ask for explicit user permission before taking any exploratory or interactive actions like scrolling and clicking. Anthropic notes the feature integrates particularly well with its recent Dispatch tool, which lets users assign tasks to the desktop app from their phone, but cautions that performance isn't perfect—complex tasks may need a second attempt, and screen-based control is slower than direct API integrations. The company is releasing it early to gather real-world feedback on its strengths and limitations.