← Back to all news

Claude Dispatch Is the First AI Tool That Works Like a Real Coworker

Send a task from your phone, come back to finished work. Anthropic ships remote control for Claude the week OpenAI is still asking users to clone a GitHub repo and configure a WebSocket gateway.

Claude Dispatch interface showing phone and desktop connection
Image via FindSkill.ai

On March 17, Anthropic launched Claude Dispatch — a research preview that lets you message Claude from your phone and pick up finished work when you return to your desk. Setup takes under two minutes: download Claude Desktop, open Cowork, click Dispatch, scan a QR code. That’s it. No API keys. No config files.

It sounds simple. It is simple. And that simplicity is the entire point.

What Dispatch actually does

Think of it as a walkie-talkie to an AI coworker sitting at your desk. Your phone sends the task. Claude runs it on your computer. You come back to finished work.

According to Anthropic’s Boris Cherny: “One persistent conversation with Claude that runs on your computer. Message it anytime from your phone. Come back to finished work.”

The key word is persistent. This isn’t starting a new conversation every time — it’s a continuous session that keeps running while you’re away from your keyboard. Early users are putting it to work on research synthesis, summarizing documents through Notion connectors, compiling reports, and organizing files. The pattern that’s emerging: Dispatch works best for anything you can describe as “summarize, find, list, retrieve, or compile.”

Important constraint: this is not cloud computing. Your desktop has to stay on and the Claude app has to stay open. Close the lid, Dispatch goes dark. It’s a remote control, not a remote server — a distinction that matters for how you use it.

The OpenClaw comparison that explains everything

To understand what Anthropic built, it helps to look at what it’s competing with.

OpenClaw — the open-source computer-control agent that accumulated 320,000 GitHub stars — offers similar functionality in theory. In practice, getting it running requires: installing Node.js 22 or newer, cloning the repository, running npm install, configuring a WebSocket gateway on port 18789, starting a daemon process, and accessing a local web dashboard. There are separate setup guides for macOS, Linux, Windows (WSL2), and Android. The setup is long enough that multiple third-party guides exist just to walk people through it.

Dispatch does the same core thing — your phone controls an AI agent on your machine — in three steps and a QR code scan. Latent Space described it bluntly: Anthropic built OpenClaw faster than OpenAI.

That speed gap is a product philosophy, not just a feature difference. Anthropic is betting that the future of AI-as-coworker looks like consumer software, not developer infrastructure.

The Claude Code angle is the most underreported story

Dispatch gets the attention, but Anthropic also quietly shipped something arguably more significant in the same week: Remote Control for Claude Code.

Claude Code is the agentic coding tool that is currently responsible for 4% of all public GitHub commits — a number growing fast enough that analysts project 20% by end of 2026. It runs in your terminal and can write, refactor, debug, and ship code autonomously across a whole codebase.

Remote Control connects that terminal session to the Claude mobile app. Start a task at your desk, then pick it up from your phone while you take a walk or join a meeting. Claude keeps running on your machine. You can check in, redirect, or approve from anywhere.

For people who are already using this workflow — writing CC instructions in Claude on their phone, then pasting them into a terminal session on their desktop — this closes the loop. You no longer have to be at your desk to supervise what’s happening. You sent the instruction from your phone. Now you can watch it execute, catch problems early, and redirect without switching contexts.

This is the shift from “AI as a tool you visit” to “AI as a coworker you delegate to.” The delegation doesn’t require your physical presence anymore.

Who this is for right now

Dispatch is a research preview, which means the reliability is honest about being early — roughly 50/50 on complex tasks according to hands-on testing. The limitations are real: one persistent thread (no parallel conversations), no notification when work finishes, and no utility when your desktop is asleep or offline.

But for people who are already living in Claude — using it for research, document workflows, and coding tasks — Dispatch extends that capability to wherever you are. It’s rolling out to Max subscribers ($100/month) first, with Pro ($20/month) following within days.

The tools that become infrastructure rarely announce themselves as such when they launch. Dispatch is a research preview today. The workflow it enables — delegate from anywhere, supervise asynchronously, return to finished work — is the shape of how knowledge work is going to run.

What AI Uni teaches about this

AI Uni’s BLD-201 (Claude Code Power User) track teaches exactly the skills this workflow requires: how to structure agentic tasks, write effective CC instructions, manage multi-step autonomous workflows, and supervise AI output effectively. The ability to delegate well — knowing how to spec a task clearly enough that an AI can run it without you watching — is the new high-leverage skill. AI-101 teaches the foundations for everyone starting from scratch.

Try 2 Free Sessions

Sources

  1. Boris Cherny (Anthropic) — Official Dispatch announcement, Threads
  2. FindSkill.ai — Claude Dispatch: Control Cowork From Your Phone (Setup + First Look)
  3. Latent Space — [AINews] Claude Cowork Dispatch: Anthropic’s Answer to OpenClaw
  4. Anthropic — Claude Code Remote Control documentation
  5. MacStories — Hands-On with Claude Dispatch for Cowork
  6. SemiAnalysis — Claude Code Is the Inflection Point (4% of GitHub commits)