Anthropic Just Gave Claude a Body. Here's What That Means for You.
Anthropic launched two new features this week, Dispatch and Computer Use, that let Claude work on your computer while you're not even there. This is a full breakdown of what these features do, how they work, and what you can do with them.
⚡ Key Takeaways
- Anthropic launched Claude Dispatch (March 17) and Computer Use (March 23, 2026) for Pro and Max subscribers.
- Dispatch lets you text Claude tasks from your phone and come back to finished work on your desktop.
- Computer Use lets Claude control your Mac's mouse, browser, and keyboard when no connector integration exists.
- Files stay on your machine and are never sent to Anthropic's servers during task execution.
- Both features are in research preview, meaning they're not perfect yet and Anthropic says so openly.
- Currently Mac-only. Windows and Linux users will have to wait.
- Both features are available in research preview for Claude Pro ($20/month) and Max ($100/month or $200/month) subscribers. Not available on Free, Team, or Enterprise plans.
What Anthropic Actually Launched This Week
Think about the last time you left your desk wishing you could just keep working. You're on the train, the task is sitting there half-finished, and there's nothing you can do until you're back at your computer. That frustration? It's basically what Anthropic built these two features to solve.
On March 17, 2026, Anthropic introduced Dispatch, a new feature inside Claude Cowork that keeps one persistent, continuous conversation between your phone and your desktop. A week later, on March 23, they layered in Computer Use, which gives Claude the ability to physically operate your Mac, clicking, scrolling, opening apps, running browsers, the whole thing, when it doesn't have a direct app connector to rely on.
Together, these two features represent something genuinely different in the AI assistant space. Most AI tools sit there and wait for you. Claude, with Dispatch and Computer Use, can now go do the work while you're not watching.
Claude Dispatch Explained: Your Phone as a Remote Control for Claude
Here's the clearest way I can describe it: Dispatch turns your Claude Desktop into a background worker and your phone into the boss's office.
Before Dispatch, Claude Cowork was powerful but desk-bound. You had to be sitting at your computer to assign tasks and monitor what Claude was doing. There was no mobile access. That just changed.
With Dispatch, you open the Claude mobile app, go to the sidebar, find Dispatch, and type a task in plain English. Something like: "In my Reports folder, summarize the Q1 spreadsheet and email the summary to my manager." Claude picks up that task on your desktop, works through it using your local files, your connectors (Gmail, Slack, Google Calendar, Notion, Dropbox, GitHub, and 30+ more), and messages you back with the result. You can be at the gym, at a client meeting, or standing in line for coffee. It doesn't matter.
What makes this genuinely different from other "AI on mobile" features is the persistence. The conversation never resets. Claude retains context from earlier tasks in the same thread, so you can ask follow-up questions or adjust the output without explaining everything from scratch. That's a meaningful shift from how most AI tools work right now.
"One persistent conversation with Claude that runs on your computer. Message it from your phone. Come back to finished work." Felix Rieseberg, Anthropic engineer, via Twitter/X, March 17, 2026
Claude Computer Use: What It Can (and Can't) Do Right Now
So what happens when Claude gets a task but doesn't have a connector for the app involved? That's where Computer Use comes in. And honestly, it's the part of this announcement that's harder to explain without sounding like science fiction.
Claude can now look at your screen, move your cursor, click buttons, type into fields, scroll through pages, and navigate your browser, basically everything a human does when they sit down at a computer. When you give it a task that requires, say, updating a dashboard in an internal tool you've built yourself (no API, no connector, just a browser window), Claude will open it and work through the interface the same way you would.
Anthropic is being refreshingly honest about the limitations here. In their own words: "Computer use is still early compared to Claude's ability to code or interact with text." Complex tasks may take multiple attempts. Screen-based navigation is slower than direct API interactions. And yes, it's currently Mac-only.
The computer use capability works on top of Dispatch. So the workflow becomes: you text Claude a task from your phone, Claude tries the smartest, most direct tool first (a connector like Slack or Google Calendar), and if no connector exists, it falls back to operating the computer directly. It's a graceful escalation ladder, which is actually smart product design.
How to Set Up Claude Dispatch in Under 5 Minutes
Setting up Dispatch is genuinely quick. Here's the exact path, laid out like I'm walking you through it over your shoulder.
Download and Update Claude Desktop
Go to claude.ai/download and install the latest version. You need the most recent build, older versions won't show the Dispatch tab. macOS is required right now. Estimated time: 2 minutes.
Install the Claude Mobile App and Update It
Download the Claude app on your iPhone or Android phone. If you already have it, make sure you update to the latest version. The Dispatch UI won't appear in older builds. Estimated time: 1 minute.
Pair Your Phone Using the QR Code
Open Claude Desktop, click on the Dispatch tab in the sidebar, and scan the QR code with your phone's camera. This links both devices to the same persistent conversation thread. Estimated time: 30 seconds.
Add Your Connectors in Desktop Settings
Before you go mobile, connect the apps you actually use: Gmail, Slack, Notion, Google Drive, GitHub, whatever fits your workflow. The more connectors you add, the less often Claude needs to fall back to computer use. Estimated time: 2 to 5 minutes depending on how many you add.
Send Your First Task From Your Phone
On your phone, go to the Claude app sidebar and tap Dispatch. Type a task that references a specific folder or file on your computer. Be specific: folder name, file name, what you want done. Vague prompts produce vague results. Estimated time: 1 minute.
One practical tip that saves a lot of frustration: name your skills and scheduled tasks with short, obvious labels on the desktop app before you leave the desk. That way, when you're on your phone later, you can trigger them with a quick message like "Run the weekly metrics task." Simple names make mobile use a lot smoother.
Real-World Use Cases for Professionals (Not Just Developers)
A lot of AI news coverage frames this kind of feature as something only developers care about. That's not the full story here. Let me give you the use cases that actually matter across different types of work.
For Knowledge Workers and Managers
You wake up, you're on your commute, and you text Claude: "Pull my emails from the past 24 hours and give me a priority list, anything flagged urgent or from my top 5 clients." By the time you sit down at your desk, the summary is waiting. No manual inbox triage. You just start on what matters.
For Developers (and This Is Where It Gets Powerful)
This is where Dispatch really shows its teeth. Claude can open your IDE while you're away, make changes to the codebase, run tests, and submit a pull request. Anthropic gave this as a specific example in their launch post. You assign the task from your phone during lunch, and the PR is up when you get back. For solo developers especially, this is like having a junior dev working in parallel.
For Freelancers and Consultants
Imagine you're in a client meeting and they ask you for a quick report on last quarter's numbers. You text Claude, point it to your data folder, and ask for a summary formatted as a PDF. By the time the meeting ends, it's done. That's not hypothetical. That's a task Dispatch handles right now.
For Researchers and Writers
Claude can search files on your desktop, analyze spreadsheets, pull deadlines from documents, and structure findings. One user example that came up in community discussions: pointing Dispatch to a folder of client contracts and asking for a deadline summary. It searches the files, analyzes the data, and returns a structured report, all locally, without the files ever leaving the machine.
Is It Safe? The Honest Security Breakdown
This is the question you should be asking. When you give an AI the ability to control your computer, the security model matters a lot. Here's what Anthropic built in, and where the real risks live.
The permission-first model means Claude asks before touching any new application. You're never surprised by Claude opening something you didn't expect. And you can stop it at any point, in the middle of a task, with a single command.
Local-first processing is a big one. Files stay on your machine. The contents of your documents are not sent to Anthropic's servers during task execution. This is confirmed in Anthropic's support documentation and matters a lot for anyone working with sensitive business data.
Prompt injection protection is baked in. Anthropic built automated scanning that monitors model activations specifically looking for prompt injection attacks, which are a common vulnerability for AI tools that have real-world access.
But here's the honest part: it's not perfect. Anthropic said so themselves. The security model is still evolving, and they actively recommend not giving Claude access to highly sensitive data during the research preview. Think twice before pointing Dispatch at folders containing financial records, medical data, or legal documents until the feature matures.
Also, the desktop must stay awake and the Claude Desktop app must remain open for Dispatch to work. This isn't just a technical detail; it's actually a design choice that prioritizes user control. Claude can't do anything without your machine being active and present.
Claude Dispatch vs. OpenClaw: What the Comparison Actually Tells Us
You can't write about Dispatch without talking about OpenClaw, the open-source competitor that's been generating wild buzz, including reports of lines around the block at Tencent's Shenzhen headquarters just to install it. Seriously. That happened.
The two tools are going after the same idea: AI agents that can operate computers autonomously. But the approach is very different, and the differences tell you a lot about who each product is for.
| Feature | Claude Dispatch | OpenClaw |
|---|---|---|
| Setup difficulty | Easy (QR code, ~5 min) | Technical (advanced config required) |
| Always-on background operation | No (desktop must be awake) | Yes (runs continuously) |
| Security guardrails | Strong (permission-first, injection scanning) | Weaker (open-source flexibility = exposure risk) |
| Data privacy | Local processing, files stay on device | Depends on configuration |
| Multi-model support | Claude only | Multiple AI models |
| Best for | Non-technical professionals wanting a reliable, safe setup | Advanced users who want flexibility and don't mind risk |
| Cost | Max plan: $100/month. Pro: $20/month (coming soon) | Free (open-source), but requires self-hosting |
The short version: OpenClaw is like a race car with no seat belts. Extremely capable, but you need to know what you're doing. Dispatch is the reliable daily driver with airbags. Less raw power right now, but a much lower chance of something going sideways.
Hidden Problems and Friction Points (From Real Users)
I want to be straight with you here, because early reviews and community posts have surfaced some real friction that the official announcement didn't highlight. Here's what you should actually know going in.
Hidden Problem 1: The Thread Can't Be Restarted
Dispatch runs on a single persistent thread. Right now, there's no way to start a fresh thread or manage multiple parallel threads. If a task goes wrong or context gets muddled, your main option is to just keep going or clear things out manually. Early users found this genuinely limiting for complex or multi-phase projects.
Workaround: Keep your task prompts modular and self-contained. Instead of one big complex instruction, break it into separate, sequential asks. Less elegant, but it avoids context pileup.
Hidden Problem 2: Connector Access Isn't Always Obvious
One reviewer on The New Stack documented an issue where Claude had access to a Google Drive web integration but not the actual local Google Drive folder, which wasn't clear from the setup UI. The task eventually worked, but only after multiple permission requests and some confusion.
Workaround: Before going mobile, do a dry run on desktop. Confirm each connector is linked correctly by giving Claude a simple test task using that connector first.
Hidden Problem 3: Complex Computer Use Tasks Sometimes Fail on First Try
Screen-based navigation is slower than API access by nature, and Anthropic acknowledges that complex tasks may need multiple attempts. One early user noted that after a string of permission requests during a more involved workflow, Claude "just gave up on the query and died," in their words.
Workaround: Start with simpler, well-defined tasks while you get a feel for what works reliably. Use computer use for tasks where a connector doesn't exist and the task is relatively linear, not for anything requiring complex multi-step navigation across several apps simultaneously.
Plan Availability: Who Can Access Dispatch and Computer Use
Both Claude Dispatch and Computer Use are available in research preview for Claude Pro and Max subscribers on macOS. They are not available on the Free plan, and Anthropic's official documentation confirms that Team and Enterprise plans are also currently excluded from this preview.
Here's the full breakdown of what each plan includes:
| Plan | Monthly Price | Dispatch Access | Computer Use | Notes |
|---|---|---|---|---|
| Free | $0 | No | No | Not included in research preview |
| Pro | $20/month | Yes | Yes | Full access, macOS only |
| Max 5x | $100/month | Yes | Yes | Full access + 5x usage limits vs Pro |
| Max 20x | $200/month | Yes | Yes | Full access + 20x usage limits vs Pro |
| Team / Enterprise | From $25/user/month | No | No | Excluded from this research preview |
The key distinction between Pro and Max is not feature access, it's usage volume. Both plans get the same Dispatch and Computer Use capabilities during this research preview. Max subscribers get 5x or 20x the usage limits of Pro, which matters most for developers running Claude Code intensively or professionals delegating large volumes of tasks throughout the day.
Since both features are in research preview, Anthropic is actively collecting feedback and pushing updates. The experience will improve over time for all subscribers on eligible plans.
Claude Dispatch & Computer Use: Complete Feature Reference
What Are Claude Dispatch and Computer Use?
Claude Dispatch is a feature inside Claude Cowork (Anthropic's desktop AI agent platform) that enables a persistent, single-threaded conversation between a user's mobile phone and their desktop computer. Released on March 17, 2026, it allows users to assign tasks to Claude via the mobile app and collect completed results on the desktop, without being physically present at the machine.
Claude Computer Use is a complementary capability released on March 23, 2026, that allows Claude to directly control a Mac computer's mouse, keyboard, browser, and screen. It activates automatically when no compatible app connector exists for a requested task. Both features are in research preview.
| Attribute | Details |
|---|---|
| Launch date (Dispatch) | March 17, 2026 |
| Launch date (Computer Use) | March 23, 2026 |
| Platform availability | macOS only (research preview) |
| Plan availability | Pro ($20/mo) and Max ($100–$200/mo) — macOS only. Free, Team, and Enterprise excluded. |
| Data storage | Local only; files do not leave the user's device |
| App connectors supported | 38+, including Gmail, Slack, Notion, GitHub, Google Drive, Figma, Trello, Asana |
| Persistence model | Single continuous thread; no multi-thread support yet |
| Security model | Permission-first; prompt injection scanning; user can stop at any time |
| Primary use cases | File analysis, email management, PR submissions, report generation, scheduled task automation |
| Limitation | Desktop must remain awake; Windows/Linux not yet supported; complex tasks may fail initially |
| Main alternative | OpenClaw (open-source, multi-model, more flexible but less secure) |
Frequently Asked Questions
Which Claude plans include Dispatch and Computer Use?
Both features are available in research preview for Claude Pro ($20/month) and Max subscribers ($100/month or $200/month) on macOS. They are not available on the Free plan. Anthropic's documentation also confirms that Team and Enterprise plans are excluded from this current research preview.
Does Claude Dispatch work on Windows or Linux?
Not yet. As of March 2026, Dispatch and Computer Use are available on macOS only. Anthropic has confirmed Windows support is coming but has not announced a specific timeline.
Does my computer need to stay on for Dispatch to work?
Yes. Your desktop must be awake and the Claude Desktop app must be open and running for Dispatch to execute tasks. If your computer goes to sleep or the app closes, Claude can't work on any assigned tasks.
Is Claude Dispatch secure for sensitive business files?
File processing happens locally and Anthropic states that document contents are not sent to their servers during task execution. However, since this is a research preview, Anthropic recommends avoiding highly sensitive data for now, think financial records, medical data, or client contracts until the feature stabilizes.
Can Claude Dispatch run tasks automatically on a schedule without me texting it?
Yes, with scheduled tasks configured in Claude Desktop. You can set up recurring tasks (like a daily email summary or weekly metrics pull) that run automatically at a set time. You can also trigger them manually from your phone with a short text command.
How is Claude Dispatch different from Claude Code's remote control feature?
Claude Code also added Dispatch support, but it's primarily aimed at developers using it for coding workflows (IDE changes, tests, PRs). Claude Cowork Dispatch is designed for a broader, less technical audience working with documents, emails, and productivity apps. The underlying technology is similar, but the interface and use case focus differ.
What's the best use case for Claude Dispatch + Computer Use for a non-developer?
The most immediately practical use case is automated morning briefings. You set up a scheduled task to pull your emails, calendar, and key files, summarize everything, and send the result to your phone before you reach the office. You arrive already knowing what your day looks like. No screen time required until you're ready.