Claude takes over the desktop: Anthropic unveils AI that can run your computer independently

In a significant leap for artificial intelligence, Anthropic has introduced a new capability that allows its AI assistant, Claude, to take full control of a user’s computer—executing tasks autonomously while the user is away.

The update marks a shift from AI as a passive tool to an active digital operator. With system-level access, Claude can now navigate screens, operate a mouse and keyboard, open applications, browse the web, and even manage workflows such as spreadsheets—tasks traditionally performed manually at a desk.

The announcement, shared on X, comes days after the rollout of “Dispatch,” a feature that enables users to control Claude remotely via their smartphones. Together, the tools position Claude as a virtual co-worker capable of completing assignments independently.

Alex Albert, a researcher at Anthropic, highlighted the potential of the update, noting that a future where users no longer need to open their laptops to get work done is “becoming real very fast.”

From assistant to autonomous agent

To function, Claude integrates with connected apps such as Slack and Calendar, while seeking permissions for additional access when required. According to Felix Riesenberg, the feature is designed to work seamlessly with Dispatch, enabling users to assign tasks remotely and let the AI execute them end-to-end.

This effectively transforms Claude into a personal AI agent—similar in concept to emerging systems like OpenClaw—capable of handling workflows without continuous human input. Users can issue instructions on the go and return to completed tasks.

Limited rollout, wider implications

For now, the feature is available as a research preview for paid users of Claude Cowork and Claude Code, and is currently limited to macOS devices. Full functionality requires both the desktop and mobile apps to be updated and paired.

The move reflects a broader industry push towards “agentic AI,” where systems are designed not just to assist but to act. Tech rivals are rapidly advancing in the same direction—Nvidia has introduced its own variant, NemoClaw, while Meta and OpenAI are also developing similar autonomous agents.

As AI begins to take on execution, not just instruction, the line between user and operator is blurring—potentially redefining how work itself gets done.