Claude Cowork Dispatch: Turning Your Phone Into a Remote Control for Desktop AI Agents

TL;DR: Dispatch in Claude Cowork (beta for Pro and Max) lets you control persistent desktop AI agents from your phone. How it works, governance, and constraints.

Why this matters: Dispatch is a capability inside Claude Cowork, not a separate standalone product. According to Anthropic's Help Center, Dispatch is available in beta for Claude Pro and Max plans, on iOS and Android. It turns your desktop into a persistent AI execution engine you can monitor and direct from your phone. For a CTO, founder, or engineering leader at a growing software team, mid-sized company, 20-person company, or professional services firm, the practical bet is whether Dispatch shifts enough sustained-task work into the agent layer to justify the operating-model change. The core idea is a deliberate architectural split: the desktop does the work, the phone gives you control.

This is not a mobile-native AI assistant. Dispatch is a remote-control layer over a desktop agent that remains running, connected to your tools, and capable of acting on your behalf while you are away from your desk. That distinction matters when evaluating it.

A few practical constraints worth naming up front. Dispatch needs the current Claude Desktop application and the current Claude mobile app installed and updated. Your desktop must be awake (not asleep or shut down), and Claude Desktop must remain open while a session is running. The computer-use tool inside Claude can act outside the Cowork sandbox: when it does, it depends on your operating-system level controls plus your own safety configuration, not on Cowork's sandbox boundary. Plan accordingly before granting broad connector or computer-use permissions.

Why This Matters Now: Context for European Engineering Teams

For engineering leads and technical founders operating across distributed teams, the practical bottleneck is rarely capability. It is continuity. An AI agent that requires you to be at your desk to initiate, monitor, or unblock work does not map onto how most senior technical people actually move through their days.

The European regulatory context adds a specific dimension. GDPR and sector-specific data governance requirements mean that any agentic tool handling documents, code, or customer-adjacent data needs a credible answer to the question: where does the data go, and who controls it? The sandboxed local execution model in Dispatch is directly relevant here. Files stay on your machine; nothing beyond the conversation is sent to Anthropic's servers. That is not a complete governance answer, but it is a meaningful architectural constraint that simplifies the compliance conversation compared to fully cloud-executed alternatives.

There is also a broader shift underway in how engineering organisations are structuring AI-assisted delivery. The move toward what practitioners are calling "two-lane stacks" (where agentic tools handle sustained background tasks while humans direct at a higher level) makes mobile-accessible control surfaces more operationally relevant. Dispatch is an early, concrete implementation of that model. Teams already evaluating how to structure coding agent workflows will find it lands in a familiar conceptual space.

How Claude Dispatch Works

Setup is straightforward. Update Claude Desktop to the current version, navigate to the Cowork section, select Dispatch, and scan the QR code using the Claude mobile app. Your phone is now paired to your desktop session.

From that point, the architecture behaves as follows: the desktop is the execution engine, and the phone is the remote control. Claude continues to run on your desktop (with access to your local filesystem, your connected tools, and any active Claude Code sessions) while the mobile app gives you a low-friction interface to issue instructions, review status, and respond to prompts without sitting down.

Push notifications handle the handoff points. When a task completes or when the agent needs a decision to proceed, you get a notification on your phone. You respond, and the desktop continues. This keeps long-running tasks moving without requiring you to babysit a terminal.

The connector coverage is broad. Dispatch ships with native integrations across tools like Notion, Gmail, Slack, Google Drive, Google Calendar, GitHub, GitLab, Figma, Todoist, Asana, Trello, and Dropbox, among others. These run through the desktop client, which means the same data-locality guarantees apply: the integrations are executed locally, not routed through a cloud intermediary. (For the current connector list and counts, check Anthropic's Help Center; the lineup expands.)

One capability worth flagging specifically: you can launch Claude Code sessions from your phone, which then run on your desktop. For engineering leads, this means you can kick off a code generation or refactoring task during a commute and have it executing on your development machine before you arrive. The risk-aware operating model for Claude Code in team contexts applies here. Dispatch does not change the underlying execution behaviour of Claude Code; it changes when and where you initiate it.

What Teams Can Actually Do With It

The following use cases reflect what the architecture makes feasible. They are illustrative, not validated case studies.

1. Kick off background research tasks before leaving for a meeting. Instruct the desktop agent to compile information from Notion, Drive, and email on a specific topic. Review the output on your phone when it is ready.

2. Initiate code generation from your phone. Launch a Claude Code session targeting a specific file or feature branch. The session runs on your desktop; you get notified when it completes or needs review.

3. Monitor long-running document workflows. Set the desktop agent to work through a batch of documents (summarising, tagging, routing) and receive push notifications at decision points rather than checking in manually.

4. Manage cross-tool coordination tasks. Use the native connectors to have the desktop agent move information between Slack, Notion, and GitHub without requiring you to be at the machine to orchestrate each step.

5. Respond to agent blockers asynchronously. When the agent hits an ambiguous instruction or reaches a permission boundary, it surfaces the question via push notification. You answer from your phone, and execution continues.

The practical value scales with the length and complexity of the tasks you delegate. For short, discrete operations, the overhead of the setup is not worth it. For multi-step tasks that run over minutes or hours, the asynchronous control model pays off.

What to Evaluate Before Adopting It

Data governance. Local execution addresses some concerns, but it does not resolve all of them. Verify that your organisation's data classification policies are compatible with the connected integrations. Tools like Gmail and Slack carry their own data handling implications regardless of where the agent runs.

Access scope. The desktop agent operates with the permissions of your local user account and your connected app credentials. Before enabling broad connector access, map what data surfaces the agent can reach and apply the principle of least privilege. This is standard practice, but it is easy to overlook during initial setup. Teams working through post-selection standardisation of coding agent access will recognise the same governance questions here.

Plan and device requirements. Dispatch is currently in beta on Claude Pro and Claude Max plans. If your team is on a Free or Team-only configuration, you will need a procurement decision before evaluating it, not just a configuration change. Confirm iOS and Android coverage against your team's actual device distribution. Dispatch also requires the current Claude Desktop application running and an awake desktop machine; an offline or sleeping laptop will pause an in-flight session.

Task suitability. Dispatch is most useful for tasks where execution time is measured in minutes or longer. It does not meaningfully improve short, synchronous interactions. Be specific about which workflows you are targeting before you build team habits around it.

Audit and observability. Understand how Dispatch logs agent actions. For regulated industries or teams with internal audit requirements, you need to know what record is kept of agent behaviour across sessions.

Operator Takeaway: What to Try This Week and What Not to Automate Yet

What this means for your day-to-day workflow. Dispatch is the cleanest available answer to "I want my agent to keep working when I am not at my desk", but the beta-on-Pro-and-Max framing matters: do not build a team-wide habit around a beta surface that may change. Treat Dispatch as a personal-productivity tool first, a team workflow second.

What to try this week (low-risk, high-signal):

1. Update Claude Desktop and the Claude mobile app, pair them via the QR-code flow, and run one long-running task end to end (a multi-document summary, a multi-file code refactor). The handoff and notification model is the load-bearing question; one round-trip will tell you whether it fits how you work.
2. Walk through the connector permission scopes that Dispatch surfaces (Notion, Gmail, Slack, GitHub, GitLab, Drive, Calendar, etc.) and apply least-privilege scopes per connector before you use Dispatch on any work-related task. The connector grants outlive any single Dispatch session.
3. If you use Claude Code, try launching one Claude Code session from your phone to your desktop. The latency and the reviewer-on-mobile UX is the operator-facing test that decides whether this changes your routine.

What not to automate yet:

• Computer-use actions that operate outside the Cowork sandbox without explicit safety controls. The Cowork sandbox is one boundary; OS-level computer-use actions (clicking buttons in a desktop app, typing into a system dialog) escape that boundary. Configure Claude Desktop's permission and approval settings before you let computer-use act unattended.
• Customer-facing workflows on a beta surface. Dispatch is currently in beta; treat it as suitable for internal-only sustained tasks until it reaches a more stable contract.
• Trusting "desktop is awake" as a given. Energy-saver, screen-sleep, and corporate MDM policies can put your machine to sleep in the middle of a Dispatch task. If you depend on Dispatch, configure a "do not sleep while plugged in" setting on the host machine and confirm Claude Desktop is allowed to keep running.

Frequently Asked Questions

What is Claude Cowork Dispatch?

Dispatch is a capability inside Claude Cowork (not a separate standalone product). It is currently available in beta on Claude Pro and Max plans. It keeps a Claude agent running persistently on your desktop and gives you a mobile interface, via the Claude app on iOS or Android, to control, direct, and receive updates from that agent when you are away from your desk.

How is Dispatch different from using Claude on mobile directly?

Using Claude directly on mobile gives you a standard conversational AI session on your phone. Dispatch is different in kind: your phone becomes a remote control for an agent running on your desktop, which has access to your local files, connected tools, and active Claude Code sessions. The execution happens on your desktop; the phone is only the control surface.

Does Claude Dispatch work with Claude Code?

Yes. You can launch Claude Code sessions from your phone, and those sessions run on your desktop. This means you can initiate code generation, refactoring, or analysis tasks remotely and have them executing on your development machine without being at your desk.

Is Claude Dispatch secure for enterprise use?

The sandboxed local execution model means files and data stay on your machine: nothing beyond the conversation is sent to Anthropic's servers. This is a meaningful architectural constraint for data locality concerns. However, enterprise suitability also depends on your specific data classification policies, the scope of permissions granted to connected integrations, your organisation's audit requirements, and your treatment of computer-use actions that operate outside the Cowork sandbox. Local execution simplifies some compliance questions but does not resolve all of them, and the beta status of Dispatch is itself a contract signal: features and behaviour can change.

Further Reading

• One Coding Agent or Two-Lane Stack? How Technical Leaders Should Decide in 2026
• Claude Code for Teams in 2026: The Risk-Aware Operating Model
• What CTOs Should Standardise First Once They Pick One Coding Agent
• What Anthropic's Claude Managed Agents Means for SME Operators
• How to Run an Internal AI Pilot Without Creating Governance Debt
• Claude Routines for Engineering Teams: What to Automate

• For strategic guidance on AI tooling adoption: Our AI Consulting services help teams evaluate and govern new agentic tools before they create operational debt.
• For a structured readiness review: Our AI Readiness Assessment identifies governance gaps before they become expensive.
• For building the delivery model: Explore our AI Development Operations services.