Anthropic has launched Dispatch, a new research preview feature within Claude Cowork that allows users to remotely interact with an AI agent running locally on their Mac from a mobile device. The feature is rolling out first to Max subscribers, with broader access expected soon.

This matters because it expands AI from a tool users interact with in real time into a system that can continue executing tasks on their computer even when they are away—introducing a new model for how work gets delegated to AI.

Dispatch works by pairing the Claude desktop app with a mobile device, allowing users to send instructions and monitor progress while the AI operates within a sandboxed environment on their machine. This builds on a growing trend toward local AI execution, where systems can directly access files, applications, and workflows.

The feature is most relevant for knowledge workers, developers, and businesses exploring how AI can automate multi-step tasks across real environments, particularly where data privacy and local control are important.

In short, Dispatch allows users to remotely assign and monitor tasks handled by a Claude-powered AI agent running directly on their own computer.

Dispatch is a remote-control layer for Claude Cowork that enables persistent, local AI agents to be accessed and directed from a mobile device.

Anthropic Dispatch is a Claude Cowork feature that enables users to remotely control a persistent AI agent running locally on their Mac, extending AI workflows beyond the desktop into continuous, real-world task execution.

Anthropic’s Dispatch builds on its Claude Cowork environment, where AI agents can already interact with files, tools, and applications on a user’s machine. With Dispatch, that same Claude Cowork session can now be accessed remotely through a paired mobile device—extending those capabilities beyond the desktop and building on recent developments such as Manus’ “My Computer” feature, which also enables AI agents to operate directly on local machines.

Setup begins with updating the Claude desktop app, where a new Dispatch option appears inside Cowork. Users then pair their mobile device by scanning a QR code, linking the active desktop session to the Claude mobile app.

Importantly, Cowork itself is not available as a native feature within the iOS app. Instead, Dispatch functions as a remote access layer. Once paired, a Dispatch entry appears in the mobile app, allowing users to reconnect to the active Cowork session running on their Mac.

From there, users can send instructions from their phone—even while away from their computer—and check in as Claude works through tasks on the Mac in the background. This allows work to continue without being physically at the machine, with users returning later to completed results. The Mac must remain powered on with the Claude app open for tasks to continue running.

Unlike cloud-based AI assistants, Dispatch relies on a local execution model, meaning the AI interacts directly with the user’s files and environment while operating within a sandboxed structure designed for control and safety.

With Dispatch, users can start tasks from anywhere and have Claude carry them out directly on their Mac, using the computer itself to complete the work.

This includes tasks such as:

Instead of handling just one step at a time, Dispatch allows Claude to work across files, tools, and apps to complete larger tasks—closer to how people actually use their computers.

The system requires the Mac to remain powered on with the Claude app open. While this means it’s not yet fully always-on, it unlocks access to local files and applications that cloud-only AI tools can’t reach today. As the feature evolves, this setup is likely to become more seamless and reliable.

It also changes how users interact with their computers. Rather than controlling the screen through remote access tools, users simply tell Claude what they want done and let it carry out the work in the background.

Early testing from MacStories suggests that Dispatch is still in an early stage, with mixed results across different types of tasks.

In practice, performance can vary. Some tasks complete as expected, while others are slower or fail altogether, particularly when interacting with more complex system-level actions. This shows up in a few key ways:

For now, this means Dispatch isn’t reliable enough to fully replace hands-on workflows when you’re away from your computer. However, it offers a clear preview of how AI agents could evolve—handling more complex, multi-step work over time as the technology improves.

Anthropic describes Dispatch as using a controlled execution model, designed to give users visibility into how AI systems interact with their computers.

This model includes several key safeguards:

Instead of giving the AI full freedom, this approach keeps the user in control—allowing them to decide what Claude can access and modify as it works.

As AI systems gain deeper access to personal and professional workflows, this balance between capability and oversight is likely to become an increasingly important part of how these tools are designed and adopted.

Q: What is Anthropic Dispatch?
A: Dispatch is a feature within Claude Cowork that allows users to remotely control a Claude-powered AI agent running on their Mac from a mobile device.

Q: How does Dispatch work?
A: Users pair their mobile device with the Claude desktop app, allowing them to send instructions and monitor tasks while the AI executes them locally on their computer.

Q: What makes Dispatch different from typical AI assistants?
A: Most AI tools operate in the cloud. Dispatch enables AI to interact directly with local files, applications, and workflows on a user’s machine.

Q: Who can use Dispatch today?
A: It is currently rolling out to Max subscribers, with Pro users expected to gain access shortly.

Q: Does Dispatch work if your computer is off?
A: No. The Mac must be powered on with the Claude app open for tasks to run.

Q: Is Dispatch reliable enough for everyday use?
A: Not yet. Early testing shows inconsistent results and slower performance, as the feature is still in a research preview stage.

Q: Is it safe to let Claude access your computer?
A: Claude runs in a sandbox environment, and users must approve actions before the AI can access files or make changes.

Q: Who benefits most from this feature?
A: Knowledge workers, developers, and businesses looking to automate workflows across local files, tools, and applications.

Anthropic’s Dispatch highlights a growing development in how AI systems are being integrated into real-world workflows, particularly as agents gain the ability to operate across devices and environments.

The key point: AI is moving from something users interact with in real time to something that can be assigned work, operate independently on local systems, and be managed remotely.

Who should care: If you are a knowledge worker or operator, this changes how tasks get done—shifting from completing work step-by-step to assigning it to an AI agent that continues running on your computer while you focus elsewhere.

If you are a builder or product leader, this aligns with parallel developments like Manus’ “My Computer” feature, where AI agents operate directly within a user’s environment. These approaches point toward a model that combines local execution with remote orchestration.

If you are focused on enterprise or security, the local-first, approval-based model provides a way to use AI with sensitive data while maintaining control over what the system can access and modify.

Why this matters now: Multiple companies are building toward the same model: AI agents that don’t just respond to prompts, but execute tasks across real systems and continue working outside of active user sessions.

What decision this affects: The decision is how quickly to adapt workflows, tools, and expectations for a world where AI agents operate continuously on your systems—shifting from task-by-task interaction to delegation, monitoring, and orchestration of work.

In short, AI is moving from a tool you use into a system that continues working on your behalf—across your devices, even when you’re not present.

The takeaway: The organizations that learn to manage AI agents as workers—not just software—will be the ones that redefine productivity in the next phase of AI adoption.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.