AI & TechArtificial IntelligenceBigTech CompaniesNewswireTechnology

Claude Code and Cowork Automate Computer Tasks

▼ Summary

– Anthropic’s Claude can now autonomously control a user’s computer to open files, use browsers, and run dev tools with no setup required.
– This feature is available as a research preview for Claude Pro and Max subscribers, but is currently limited to macOS devices.
– It builds on prior autonomous capabilities and extends them to Claude’s Code and Cowork AI agents designed for programmers.
– Access requires the Claude desktop app on macOS paired with the mobile app, and Claude always asks for explicit permission before performing actions.
– The feature works well with the cross-device Dispatch tool but may need retries for complex tasks and is slower than direct software integrations.

Anthropic has released a significant update enabling its Claude AI to directly operate a user’s computer. This new functionality, available in the Claude Code and Cowork tools, allows the AI to autonomously open files, control web browsers, and run development applications. The company states this requires no initial setup and can function even when the user is not physically present at their machine.

Currently offered as a research preview, the feature is accessible to Claude Pro and Team subscribers. It is initially limited to the macOS operating system. This advancement expands upon the autonomous capabilities first introduced with the Claude 3.5 Sonnet model, now specifically applied to the programming-focused Code and Cowork AI agents.

To use the feature, the Claude desktop application must be installed and active on a compatible Mac. It also needs to be paired with the Claude mobile app. The system is designed to first utilize direct integrations with supported services like Slack and Google Workspace. When a dedicated connector is unavailable, Claude will instead control the computer directly by manipulating the browser, mouse, keyboard, and screen. A key privacy and security measure is that the AI will always ask for explicit permission before it begins exploring, scrolling, or clicking to accomplish a given task.

The company notes this automation works “especially well” with Dispatch, a recent cross-device feature that lets users assign jobs to the desktop Claude app from their phone. However, Anthropic is transparent about current limitations. Executing actions through the screen is slower than using a direct API connection, and complex assignments may require a second attempt. The early release is intentional, as the company aims to gather user feedback on where the technology excels and where it needs further refinement.

(Source: The Verge)

Topics

claude ai update 98% autonomous computer control 96% ai programming tools 94% macos integration 92% claude pro subscription 90% cross-device task assignment 88% service connectors 86% user permission protocol 84% research preview release 82% desktop-mobile pairing 80%