On March 23, 2026, the gap between an AI that answers questions and an AI that completes the work became much smaller. Anthropic announced that day that Claude can now take control of a user’s Mac and execute tasks autonomously, directly responding to the wave that swept the industry following the viral rise of the open-source agent framework OpenClaw.
Users can now send Claude an instruction from their phone, and the AI agent will then complete the task on their computer. After receiving a task, Claude can open applications on the computer, navigate through the web browser and fill in spreadsheets. Anthropic demonstrated the feature with a video where a user running late for a meeting asks Claude to export a presentation as a PDF and attach it to a calendar invitation, and the AI does it without manual input.
When Claude receives a task, he checks whether it has the appropriate integrations to complete it through tools like Google Calendar and Slack. If it doesn’t have the right connector, it falls back to controlling the computer as a human would, using the screen to navigate. It can open files and documents, manage a web browser, and run development tools autonomously.
Integration with Claude Code is especially relevant for developers. Claude can make changes within the integrated development environment, submit pull requests, run tests and more, leaving the user free for other tasks. The system works on the principle of “permission before action”: Claude asks for permission before accessing each new application, users can stop the task at any time, and sensitive applications such as investment platforms and crypto wallets are blocked by default.
The current limitations are clear. The feature currently only works with Mac computers, while Windows and Linux users have to wait, and screen-based operations are significantly slower than direct API integrations. Anthropic plans to bring Windows x64 support in future updates. The feature is available as an exploratory preview for Claude Pro and Max subscribers, with the company warning that it is “still early compared to Claude’s ability to code or interact with text” and that bugs remain a possibility as safeguards are developed.
The combination of computer control and the Dispatch function, which allows users to continuously talk to Claude from a phone or desktop computer and assign agent tasks, opens a new level of automation: from the morning briefing to the automatic launch of tests. At a time when OpenAI, Google and Perplexity are chasing the same goal, on March 24, 2026, Anthropic delivered the most concrete consumer answer to the question of what it means when AI not only talks to you, but works for you, reports CNET.