Claude Dispatch and the Laptop That Can’t Sleep

My laptop isn’t allowed to sleep anymore. That’s the deal I made with Claude Dispatch, the feature Anthropic shipped on March 17, 2026 inside its desktop app, Claude Cowork, that lets me hand my computer a job from my phone while I’m out.^[1] Send Claude a task on the train, get a finished spreadsheet back before I’ve found a seat at the coffee shop. Genuinely handy, when it works. But there’s one requirement in the setup screen that gives the game away.

The requirement: your computer has to stay awake, with the Claude app open, the entire time. If it sleeps, Dispatch dies.^[2] Most write-ups file that under “annoying beta limitation, they’ll fix it.” I don’t think they will, at least not soon. The awake laptop isn’t a rough edge Anthropic forgot to sand down. It’s the direct result of a choice about where the work runs.

Here’s the part almost every explainer gets wrong. They say Dispatch “runs locally on your computer.” That’s half true, and it’s the wrong half. The model, the actual Claude doing the thinking, runs in Anthropic’s cloud, the same as it always has. Your phone sends an instruction, Anthropic’s servers work out what to do, and then they reach down into your laptop to do it: open the file, read the spreadsheet, drive the browser you’re already logged into.^[3] The thinking is remote. Only the hands are local.

That split is easier to see as a picture.

It’s a research preview, and it shows

Before the strategy read, the receipts. Dispatch went out as a research preview, Anthropic’s label for software it’ll ship but keep breaking, macOS only, to Max subscribers first and Pro a couple of days later.^[1] It costs $20 a month on Pro and $100 or $200 on the two Max tiers.^[4,5] It runs one task at a time, in a single conversation you can’t split into branches.^[2] It joins the run of separate products Anthropic pushed out this spring, including Claude Design, which landed a month later on April 17. And it isn’t reliable yet.

John Voorhees at MacStories ran Dispatch through twelve real tasks the week it launched. It finished six.^[6] Another tester ran his own trials and landed on the same number.^[7] The pattern is consistent: ask Dispatch to find and summarize something and it usually works; ask it to act across apps, send the file, open Shortcuts, post the message, and it’s a coin flip. That’s before you factor in Claude’s own rough patches of downtime.

The good reason to keep the hands local

Let me argue the other side for a second, because the local design isn’t lazy. It’s the opposite. A cloud agent boots a blank virtual machine somewhere in a data center. It has no idea what’s in your Downloads folder, it isn’t signed into your email, your Notion isn’t connected. To do real work it needs you to upload and re-authenticate everything first. Dispatch skips all of that. It runs where you already live, so it can grab the contract you saved last Tuesday without you uploading a thing. Anthropic’s own description of Cowork leans on exactly this: it runs “where most knowledge work is done: in local files, folders, and the applications people use every day.”^[12]

The advantage comes welded to a constraint. To reach your real stuff, Claude has to run on the machine that holds it, and that machine has to be awake. You also hand it a lot of trust. Anthropic warns that Claude can see whatever is on your approved screens, “including personal data, sensitive documents, or private information,” which is why one reporter’s IT department made her run the whole experiment on a throwaway laptop.^[8]

Everyone else already moved the hands to the cloud

This is where the half-step shows. Dispatch is the only serious agent that makes you keep a machine awake. The rest run in a data center and keep working with your laptop shut.^[10,11]

The clearest tell is Anthropic’s own lineup. Claude Code on the web, the cloud version of its coding agent, runs on Anthropic-managed machines; you close your laptop and it keeps going.^[9] The cloud routines it shipped earlier this year run on its servers too, on a schedule, nothing open on your desk. So Anthropic plainly knows how to build the always-on version. It ships one for code. Dispatch is the same idea pointed at your files and your apps, except those still live on your laptop, so the agent is stuck there with them. I think Dispatch is a placeholder for a product that doesn’t exist yet: a cloud agent that can reach into your real desktop without that desktop being on. The awake-laptop rule is the seam where the missing piece shows through.

So, should you use it?

If your tasks are the find-and-summarize kind, and you don’t mind leaving a machine awake, it’s a real time-saver today. The wins testers report are exactly those: email triage, digging a file out of a folder you can’t remember the name of, pulling a number from a spreadsheet.^[6] Ask it to act across apps and it’s a coin flip, so check the result. And if your dream is to fire off a job and snap your laptop shut, that isn’t this product. That’s the next one.

The brain is already in Anthropic’s cloud. The hands are on your desk, and they’re not allowed to sleep. The day they move to the cloud too, the day you can send Dispatch a task and close your laptop, is the day it stops being a preview and becomes the product it’s been gesturing at all along. Until then, the server is the laptop on your desk. The server is you.