Claude controls your computer, and you assign tasks from your phone. This is the new reality of working with AI

Artificial intelligence is gradually becoming not just a tool for answering questions, but a true assistant capable of performing work for you. Anthropic has just announced that its chatbot Claude has learned to directly control a user’s computer. Thanks to new features in the Claude Cowork and Claude Code applications, AI can click the mouse and type on the keyboard to complete assigned tasks. In conjunction with the Dispatch feature, you can also assign work from your mobile phone, no matter where you are – Claude will perform it on your computer, even if you’re not currently at it.

Claude sits at the computer so you don’t have to

When performing tasks, Claude first reaches for the most precise available tool. If it has direct connectors to services like Slack or Google Calendar, it will use them. However, if such a connector does not exist, it takes control of the browser, mouse, keyboard, and screen. It can scroll, click, open files, and explore content as needed.

In practice, this means Claude can, for example, export a presentation to PDF and upload it to a calendar, or batch edit the format of hundreds of photos – which Anthropic directly demonstrated in a video preview. Before accessing any new application, Claude will always request explicit permission, and the user can stop the activity at any time.

Be just a dispatcher on your phone

Last week, Anthropic introduced the Dispatch feature, which is now available in both Claude Cowork and Claude Code. It allows you to have one continuous conversation with Claude from both your phone and computer. You can assign a task on your mobile, attend to other things, and then open the completed work on your computer.

In conjunction with computer control, Dispatch is even more useful. Claude can work on your computer even when you’re not home. For example, it can prepare your morning briefing while you’re on the train to work. Or it can make changes in an IDE, run tests, and create a pull request. Anthropic also mentions the possibility of letting Claude follow a plan on a longer project, such as 3D printing.

Security Measures

Anthropic emphasizes that the feature was developed with a number of security measures. The system automatically scans activations within the model and detects potential prompt injection attacks (i.e., when an attacker hides a dangerous instruction for the AI within the content it is supposed to analyze). Some applications are disabled by default, and the company recommends starting with applications you trust and not working with sensitive data.

Claude controls your computer, and you assign tasks from your phone. This is the new reality of working with AI Adam Kurfürst News Adam Kurfürst News

Anthropic notes that computer control is still in an early stage of development compared to Claude’s capabilities in coding or text processing. The AI may make mistakes, and complex tasks may occasionally require a second attempt. The company is making the feature available early on purpose – it wants to learn from real-world usage and find out where it works and where it has shortcomings.

Availability and Limitations

Computer control is currently available for subscribers of Claude Pro and Claude Max plans. For now, it works only on macOS – Anthropic has not yet announced support for Windows or Linux. To activate, you need to enable the feature in the desktop application settings and ensure that the application runs in the background.

To use Dispatch, you need to pair the desktop application with the mobile one. Then you can try assigning a task from your phone and let Claude work on your computer.

Would you use the ability to control your computer with Claude from your phone?

Source: Anthropic