Claude Dispatch: Anthropic Redefines Agentic AI

A Video That Propels Claude Dispatch into the Spotlight

On March 23, 2026, a video published on the social network X catapulted the Dispatch feature of Claude into the limelight. This innovation, integrated into Cowork, allows the artificial intelligence developed by Anthropic to operate autonomously on a computer while receiving remote instructions from a mobile phone. Although it was quietly launched a few days earlier, this feature is now at the center of media attention, particularly due to the security precautions highlighted by Anthropic.

The video quickly captured public interest, reaching over 25 million views just nine hours after being posted. It shows Claude executing various tasks on a computer while its user sends commands from their smartphone, even while on the go. This demonstration generated considerable excitement, revealing the potential of Dispatch to transform the way users interact with their devices.

Gradual Launch and Growing Adoption

The Dispatch feature was initially launched in research preview mode six days before the video was published. It was first made available to Max subscribers, before being extended to Pro subscribers the following day. While technically in line with initial announcements, the enthusiasm surrounding Dispatch reflects the magnitude of the change it promises to bring to the field of agentic artificial intelligence.

With Dispatch, Anthropic is embarking on a bold path to democratize the use of agentic AI. Unlike some of its competitors, such as OpenClaw, which require advanced technical skills for deployment, Anthropic offers an accessible, ready-to-use solution.

How Claude Dispatch Works and Activation

Dispatch is a component of Cowork that relies on the interaction between two connected applications: Claude Desktop, which must be installed on your computer (compatible with macOS or Windows), and the Claude app on your phone. Both applications must be up to date to function properly.

Once the feature is activated, a unique and persistent conversation thread is established between the two devices. You can send instructions from anywhere, and Claude executes them directly on your desktop. Possible tasks include compiling reports from local spreadsheets, researching and drafting briefings from your emails, organizing folders, or creating presentations from files stored on Google Drive.

For Dispatch to work, your computer must remain on, and the Desktop application must be open, as no processing is done in the background on the cloud.

Cybersecurity: A Crucial Issue

In its documentation, Anthropic emphasizes the importance of security. Indeed, allowing a mobile agent to control a desktop agent creates a potentially vulnerable chain. A malicious instruction, manipulated content, or even a simple phishing link opened in the browser can lead to cascading actions that are difficult to reverse.

Users must be aware that everything they have allowed Claude to manipulate becomes accessible from their phone. Anthropic therefore recommends meeting three conditions before activating this feature:

• Trust each application and service in the chain.
• Understand which files and accounts are accessible.
• Know how to quickly disconnect or revoke access.

In summary, Anthropic encourages users to think not only about what Claude is supposed to do but also about what it could potentially do. This approach is based on a fundamental principle of cybersecurity: the attack surface includes everything that has been made accessible, not just what is actively used.

For Anthropic, this warning is also a way to prevent the issues faced by OpenClaw, which has often been involved in incidents related to agentic AI. For users, this underscores the importance of cybersecurity as an essential competitive advantage to fully leverage the new possibilities offered by agentic AI.