Anthropic dropped Dispatch last week. March 20, “research preview” label, the usual careful rollout. But the feature itself is anything but careful: you can now control Claude Cowork from your phone.
Sound familiar? OpenClaw has been doing the messenger-first agent thing since it blew up earlier this year. Anthropic just walked straight into their yard.
I spent the past few days testing Dispatch and comparing the two. Here’s where I landed.
What Dispatch Actually Does
You open a persistent thread on your phone. Type what you want done. Claude does it on your desktop computer.
Not on a server somewhere. On your actual machine. Claude opens apps, navigates browsers, fills out spreadsheets, pulls together documents — all running locally on the computer sitting on your desk (or kitchen table, or wherever you left it).
You can assign tasks like: summarize the work I did today, pull together a brief from these three documents, compare these two campaign reports, produce a weekly performance summary. Claude takes it from there.
The catch is that your computer needs to stay awake and the Claude desktop app has to remain open. You’re not controlling a cloud instance. You’re remote-controlling your own machine through Claude as the middleman.
It’s available to Pro and Max subscribers, with Max users getting access first. Anthropic is clearly testing the waters before a wider rollout.
The OpenClaw Comparison
OpenClaw went viral with 157K+ GitHub stars. Jensen Huang called it “definitely the next ChatGPT.” OpenAI hired its creator, Peter Steinberger. Nvidia built an enterprise version called NemoClaw.
That’s a lot of validation for one open-source project.
OpenClaw works through WhatsApp, Telegram, Slack, Signal — the apps you already have open all day. It connects to multiple AI models, not just one. And because it’s open-source, people have been building custom integrations for everything from CRM workflows to podcast production pipelines.
Dispatch takes a different approach. It’s locked to Claude, runs through Anthropic’s own app, and focuses on deep computer control rather than messenger flexibility.
Here’s the real difference: OpenClaw is a messenger-native agent that can do computer things. Dispatch is a computer-native agent you can talk to from your phone. Keep that distinction in mind. It explains almost every tradeoff between the two.
Why Marketers Should Care
Forbes nailed it: the agent market is already pushing toward mobile-first, messenger-first AI interaction. CNBC says the same thing from a different angle, reporting on the wider push from AI companies to build agents that carry out tasks without babysitting.
The shift is obvious. We’re going from “open a tool and do the work” to “tell an agent what you want and check back later.”
For marketers specifically, this changes the daily workflow. You’re at lunch and remember you need a competitor analysis for tomorrow’s meeting. You pull out your phone, tell Claude to pull the data, compare it against last quarter, and drop a summary into a doc. By the time you’re back at your desk, it’s done.
Or you’re on the train home and realize you forgot to update the campaign dashboard. You fire off a message to Claude, and it opens the spreadsheet, pulls the latest numbers, and formats the report.
This is why both Dispatch and OpenClaw matter. Your commute, your lunch break, your kid’s soccer game — all suddenly usable work time if you want it to be.
The Marketing Use Cases
Where does Dispatch actually fit in a vibe marketing workflow? I’ve been testing it. Here’s what works.
Content production. Tell Claude to open your CMS, draft three social posts based on your latest blog article, and save them as drafts. Come back and review when you’re ready. Pair this with any of the AI marketing tools in our directory and you’ve got a solid content pipeline running while you do other things.
Report generation. Point Claude at your analytics dashboard, have it screenshot key metrics, pull them into a formatted report, and save the file to your desktop. Weekly reporting that used to take an hour now takes a text message.
Research and briefs. Need competitive intel before a strategy call? Tell Claude to check five competitor websites, grab their latest messaging, and put together a comparison doc. This is where Dispatch’s browser control really shines — it can actually navigate sites and extract what it finds.
Campaign setup. If your tools have desktop apps or web interfaces, Claude can interact with them directly. Fill out forms, configure settings, copy content between platforms. It’s clunky compared to API integrations, but it works with tools that don’t have APIs at all.
These same use cases work with Claude Code for the technical side, and you can extend the capabilities further with MCP servers that connect Claude to specific marketing platforms.
The Catch
Anthropic isn’t pretending this is finished. They’re calling it a research preview for a reason.
“Computer use is still early,” they say in the announcement. “Claude can make mistakes.” That’s corporate-speak for: don’t let this thing run your ad spend unsupervised.
The practical limitations are real. Your computer has to stay awake. The desktop app has to stay open. If your machine goes to sleep or the app crashes, your remote session is dead. You’re also trusting Claude with screen-level access to your machine, which is a legitimate security consideration even though the local-first execution model means your files never leave your computer.
And compared to OpenClaw, the messenger situation is weak. Dispatch only works through Anthropic’s own interface. You can’t text Claude on WhatsApp. You can’t drop a task in your team’s Slack channel and have it picked up automatically. OpenClaw does all of that, and it does it across multiple AI providers.
OpenClaw is also open-source. You can self-host it, modify it, build on top of it. Dispatch is a closed feature inside a subscription product. For teams that want control over their agent infrastructure, that’s a significant gap.
My Take
Is Dispatch the OpenClaw killer? No. Not yet.
But Anthropic clearly sees where things are going. Sitting at your computer typing prompts is the old model. Firing off tasks from wherever you are and trusting the agent to handle execution is the new one.
OpenClaw wins right now on three fronts: messenger integration (WhatsApp, Telegram, Slack, Signal), model flexibility (use whatever AI you want), and open-source freedom. That’s a strong position, and the community momentum is massive.
Dispatch wins on two things that might matter more in the long run: deep computer control and local-first security. Anthropic’s agent can do things on your actual machine that OpenClaw’s messenger-based approach struggles with. And for any company worried about data leaving their environment, local execution is a big deal.
My prediction: these two approaches converge within six months. Anthropic adds messenger integrations to Dispatch. OpenClaw improves its computer control capabilities. The gap narrows until the real differentiator is just which AI model you trust most with your work.
For now, if you’re a marketer who lives in messaging apps and wants flexibility, OpenClaw is still the move. If you want tighter computer control and you’re already in the Claude ecosystem, Dispatch is worth trying.
Either way, assigning marketing tasks from your phone and having an AI agent execute them on a real computer? That’s not a future prediction. That’s this week.