Claude Dispatch Just Launched. Here Is What It Actually Is (And What It Is Not).

Anthropic quietly released a new feature yesterday called Dispatch — a research preview inside Claude Cowork that lets you control your Mac from your phone through the Claude mobile app.

The internet immediately compared it to JARVIS. The reality is both less dramatic and more interesting than that.

We run AI automation systems in production every day. We have opinions about where Dispatch fits, what it changes, and what it means for the ongoing Claude Code vs OpenClaw conversation. Here is the honest breakdown.

Dispatch is a mobile remote control for Cowork — Anthropic's desktop agent product.

Here is how it works: you open the Claude Desktop app on your Mac, go to Cowork, and click "Dispatch" in the left panel. A QR code appears. You scan it with the Claude mobile app on your phone. Now your phone is paired to your Mac's Cowork session.

From your phone, you can assign tasks to Claude that execute on your desktop — pulling data from local spreadsheets, searching Slack messages, building presentations from Google Drive files, organizing folders. Claude uses everything on your Mac: your files, your browser, your connected services.

The key requirement: your Mac must stay awake with Claude Desktop open and connected to the internet. Dispatch maintains one persistent conversation that follows you between devices. Start a task from your phone at lunch, come back to your desk and the results are waiting.

It is available now for Max subscribers ($100-200/month) and coming to Pro ($20/month) within days.

Dispatch is not an autonomous agent framework. It is not a competitor to OpenClaw in the way some headlines suggest. It does not run 24/7. It does not connect to your messaging apps. It does not respond to events proactively.

It is a remote control. A very good concept for a remote control — but a remote control nonetheless.

The distinction matters because the AI agent conversation keeps conflating three very different things:

• Chat interfaces — you type, AI responds (Claude.ai, ChatGPT)
• Desktop agents — AI acts on your computer with your supervision (Cowork, Dispatch)
• Autonomous agents — AI runs 24/7, monitors events, takes action on schedules (OpenClaw, Claude Code with scheduled tasks)

Dispatch moves Cowork from "desktop agent" to "desktop agent you can trigger from your phone." That is genuinely useful. But it is not the same as an always-on autonomous system.

Early hands-on testing from independent reviewers found Dispatch succeeds roughly half the time. It is slow. It cannot reliably open applications, send files via iMessage, access Terminal sessions, or retrieve active browser tabs. There are no task completion notifications — you have to check back manually to see if your task finished.

This is consistent with what "research preview" means. It is not production-grade. It is Anthropic showing the direction and collecting feedback.

For comparison: Claude Code's scheduled tasks, which we use for daily automation, have been reliable in production for months. Dispatch is not there yet. But the concept — assign tasks to your computer from anywhere — is the right one.

Anthropic now has three distinct agent surfaces:

Claude Code — lives in your terminal. Purpose-built for developers and power users. Skill system, MCP integrations, scheduled tasks via Desktop, full filesystem access. This is the production automation engine.

Claude Cowork — lives on your desktop as a GUI. Designed for non-developers who want Claude to act on their computer. Connectors and plugins instead of MCP servers. Sandboxed file access.

Claude Dispatch — lives on your phone as a remote control for Cowork. Lets you assign tasks to your Mac from anywhere. Same capabilities as Cowork, but initiated from mobile.

For developers building business automation, the answer is still Claude Code. Dispatch is exciting for the Cowork audience — people who want a desktop productivity assistant they can ping from their phone. But if you need scheduled pipelines, custom skills, and programmatic integrations, Code is the tool.

The interesting question is what happens when Dispatch matures. If Anthropic adds scheduling, notifications, and better reliability to Dispatch, it starts to look like a consumer-friendly version of what we build with Claude Code today.

The comparison everyone wants to make is Dispatch vs OpenClaw. But they solve different problems:

OpenClaw is a self-hosted, always-on autonomous agent. It connects to your messaging apps (WhatsApp, Telegram, Discord, Slack) and runs 24/7. It can monitor events, respond proactively, and execute tasks without human initiation. It is a background process that never sleeps. The tradeoff: bring your own API key (token costs add up fast), significant security surface area (a critical RCE vulnerability was disclosed recently with 135,000+ exposed instances), and the wandering problem we wrote about in our last post.

Dispatch is a human-initiated remote control. You open your phone, assign a task, Claude executes it on your Mac. No monitoring. No proactive actions. No 24/7 runtime. The tradeoff: it is sandboxed and safe, predictable subscription cost, but it requires your Mac to be awake and can only do what you explicitly ask.

For personal productivity — "organize my downloads folder while I am at lunch" — Dispatch wins on simplicity and safety.

For always-on business automation — "monitor my inbox, score leads, draft outreach, queue content" — neither Dispatch nor OpenClaw is the right answer today. Claude Code with scheduled tasks is.

The honest take: Dispatch is a preview of where consumer AI agents are headed. OpenClaw is a preview of where autonomous agents are headed. Neither is production-ready for serious business automation in the way that Claude Code already is.

If you are building AI automation for a business today, here is the practical takeaway:

Use Claude Code for production automation. Scheduled tasks, skill system, MCP integrations, predictable costs. It works. We run it daily.

Watch Dispatch for the future. The concept of assigning tasks to your computer from your phone is powerful. When Anthropic adds scheduling, notifications, and cross-device reliability, this becomes interesting for non-developer operators who want automation without touching a terminal.

Be cautious with OpenClaw for business use. The token economics and reliability issues we experienced have not changed. The security concerns are real. If you have DevOps capacity to manage it and personal use cases that fit, it can be valuable. For client-facing business automation, the risk profile is too high.

The trend is clear: every major AI company is building toward agents that act on your behalf, on your machine, with your data staying local. Dispatch, Claude Code scheduled tasks, OpenClaw — they are all pointing at the same future. The question is which one is ready for production today.

Our answer: Claude Code. Not because Dispatch or OpenClaw are bad — but because reliability and cost predictability are non-negotiable for business automation, and Claude Code is the only one that delivers both right now.

At Early to AI, we build AI automation systems for businesses using the tools that actually work in production — not the ones that make the best headlines.

If you are trying to figure out which agent framework fits your business, or you are stuck between Dispatch, OpenClaw, and Claude Code, we can help. We have tried all of them. We know what works and what does not.

No slides. Just production systems and a conversation about what yours could look like.