That Cloud Task You Started on Your Phone Secretly Depends on the Desktop You Left Running

You schedule a task to run every morning at 7am from your phone: Claude reads a local folder on your desktop, compiles a report, and emails it to you. Week one works fine. Then one evening you shut down the desktop to save power, and the next morning's 7am task "runs" on schedule — except the report comes back empty, or riddled with strange errors.

The problem isn't the schedule itself. It's an easily-overlooked precondition: any cloud task that touches local data is quietly depending, the whole time, on a desktop machine running the Desktop app.

A cloud session and local capability are two separate things

The feature gap across Cowork's web, desktop, and mobile platforms is often summarized as "mobile does less," but the real variable isn't what a given platform can do on its own — it's who is actually touching your local files. Whether you start a task from web or from the mobile app, if that task needs to read or write local files, use a browser, or control your computer, the Desktop app is always the one actually doing it. Web and mobile just "borrow" that capability through it.

That makes desktop more than just one of three platforms — it's an invisible bridge. Even when you're on your phone and the session is running in the cloud, if any local capability is involved, that bridge has to stay open. The official wording is blunt: a cloud session accessing local files requires "the desktop app to be open on that computer"; close it, and "the session keeps running but can't reach your local files."

"Keeps running" is the phrase most likely to be misread

Most people read "keeps running" and relax, assuming the task at least won't fail outright. What it actually means is that the task won't error out and stop — it will quietly continue without the local data, either falling back to stale cached data, silently skipping that step, or producing a report that looks complete but is missing something important. For a Scheduled Task, this kind of silent, error-free failure is far more dangerous than a hard error, because you won't notice it right away.

Another easily-missed detail is the platform split on Live Artifacts: desktop supports the legacy Live Artifacts format created before August 19, 2026, while web only supports the newer format created after that date — the two aren't interchangeable. If a scheduled task produces or depends on an Artifact, mixing platforms can leave you seeing inconsistent results across devices, not because the task failed, but because the versions simply don't match.

How to design Scheduled Tasks so this dependency doesn't bite

If a scheduled task relies on a local folder, browser use, or computer control, the practical move is to manage "keep the desktop on, keep the Desktop app running" as part of the same commitment as the task itself — the same way you'd remember to schedule a maintenance window for a server, treat this desktop as a "do not shut down" item on your mental checklist. If you can't guarantee the machine will be on when the task runs — overnight, or while you're traveling — the safer fix is to switch the data source to a cloud connector or a cloud folder instead of a local path, so the task no longer depends on any one specific machine being awake.

How This Affects Your Work

Projects built on a local folder have the same restriction: they only support Cowork sessions on desktop. If a scheduled task is tied to a project like that, you're stacking two layers of local dependency — the folder itself doesn't sync across devices, and neither does the execution environment. That combination is especially prone to breaking without warning the moment you switch devices. The practical check is simple: before handing any scheduled task to Cowork, ask whether its data source depends on one specific machine being powered on right now — if it does, that machine is the real single point of failure in this schedule, and deserves to be taken as seriously as the data itself.