Anthropic Expands Claude Beyond Desktop with ‘Dispatch’ Cross-Device Control Feature

Anthropic has introduced Dispatch, a new capability within its Cowork environment that allows users to manage desktop-based AI tasks directly from their smartphones, signalling a step towards continuous, cross-device AI interaction.

The feature creates a single, persistent conversation with Claude that can be accessed from both desktop and mobile. Users can assign tasks remotely and return later to completed outputs, which are processed on their computer using locally available files, connectors and plugins.

Dispatch is being rolled out as a research preview, initially to Max subscribers, with availability for Pro users expected shortly. It requires both the Claude desktop and mobile applications, alongside an active internet connection. Crucially, the desktop application must remain open and the computer awake for tasks to run.

Unlike traditional AI sessions that reset between interactions, Dispatch maintains a continuous thread. This allows users to begin a task on one device and continue or review it on another without losing context. Claude executes requests on the desktop and delivers finished outputs—such as reports, presentations or data summaries—rather than exposing intermediate steps.

The system is designed to be simple to activate, with users pairing their devices and enabling access to local files. Once configured, the experience is synchronised automatically across devices, effectively turning Claude into a persistent assistant embedded within a user’s personal computing environment.

The functionality opens up a range of use cases, from analysing spreadsheets stored locally to compiling briefings from emails and workplace tools. It also enables file management tasks that would typically require direct desktop access.

However, Anthropic has underscored the potential risks. Extending desktop-level access to a mobile-controlled AI agent creates a chain of actions that can affect files, connected services and browser activity. Errors or malicious inputs could trigger unintended consequences, particularly if users grant broad permissions.

As an early-stage release, Dispatch comes with limitations. It supports only a single conversation thread, lacks notifications when tasks are completed, and does not allow scheduled actions within the same workflow. Despite these constraints, it offers a glimpse into how AI systems may evolve into always-on, device-spanning assistants.