Claude can now control your computer, with a new feature in Code and Cowork

Anthropic has taken another step toward a truly autonomous AI agent: Claude can now control your computer directly, with no extra setup and without the user having to sit at the screen the whole time.

The new feature appears in two of the company's tools, Claude Code and Claude Cowork. With it, the model can open files on its own, launch a web browser, interact with applications and carry out developer tasks. Anthropic stresses that all of this works "without any setup": the user only needs to grant permission, and Claude gets to work.

The feature is available as a research preview for Claude Pro and Max subscribers. At launch, only macOS is supported, and support for other platforms has not yet been announced. Before it starts controlling the computer, Claude asks for the user's explicit consent, which Anthropic treats as a key safety point.

Technically, the release builds on capabilities introduced with the Claude 3.5 Sonnet model in 2024, when the company first presented "computer use": the ability of an AI to see the screen and act on an interface. Those capabilities have now been moved into the company's working products and are noticeably closer to real-world use.

The main difference from earlier iterations is integration directly into tools for development and collaboration. That turns Claude from a chatbot with curious experimental features into an agent able to complete full tasks in a working environment, for example running tests, opening a given file in an editor or loading a page in a browser to check the result.

Autonomous computer control remains one of the most discussed and contentious areas in AI agent development. Anthropic's competitors, OpenAI with Operator and Google with Project Mariner, have already introduced similar approaches. The arrival of this feature in Claude Code signals that the race for "agentic" AI is moving out of the demo stage and into products for end users.