Will it run?
Models

Anthropic adds background computer control with Cowork and Code

By Marco Vane Clawpit staff
Anthropic adds background computer control with Cowork and Code

Anthropic announced a new capability that lets its Claude assistant operate a user’s computer in the background—clicking, typing and launching applications as a human would—while the user works on other tasks. The feature is currently available in beta for Pro and Max subscribers only, and only through the macOS desktop app.

Two modes, one infrastructure

The offering includes two separate modes: Cowork for general tasks and Code for development work. Both rely on the same “computer use” infrastructure that Anthropic previously demonstrated, but this iteration runs in the background without occupying the screen or intercepting the user’s input. This marks a shift from the earlier model that required a dedicated window and close supervision.

Activation and safety

Activation is configured via the General → Computer Use settings. Anthropic supplied a safety guide that outlines how to use the feature without exposing sensitive data or enabling irreversible actions. The company has not released official performance metrics for the new version, making it difficult to gauge its accuracy against prior releases or competitors.

Platform limits and pricing

At launch the feature is limited to macOS, leaving Windows and Linux users out of the beta with no announced timeline for expansion. Within Apple’s ecosystem the capability is gated behind a higher-tier subscription, restricting access to a relatively small pool of testers.

Technical gaps

Running background automation while the user remains active requires fine-grained permission handling, process isolation and rollback mechanisms—details Anthropic did not disclose in the announcement. Without transparency on the architecture, assessing the true risk compared with the safety guide’s claims is challenging.

Competitive context

The rollout arrives as several labs are developing similar computer-agent technologies, most of which remain in closed-demo or partner-only phases. Anthropic’s current edge is immediate availability to paying customers, albeit in beta and with significant platform constraints.