If you’ve been putting off automating anything with Claude because “well, my laptop has to be on for it to run” — that excuse is gone. Anthropic has rolled out cloud-based scheduling, officially called Routines, that let Claude fire off recurring or one-off tasks entirely on Anthropic’s infrastructure. No laptop lid propped open, no “keep computer awake” setting, no missed runs because your machine went to sleep at the wrong moment.
This matters more than it might sound like on the surface. Scheduled automation only earns its keep if it’s reliable, and reliability was the weak point until now. Let’s get into what’s changed, what it’s good for, and where the old local-only approach still wins.
The Problem With the Old Way
Claude Code has offered scheduled tasks for a while, and they work fine — provided your machine is switched on, the Desktop app is running, and your computer hasn’t decided to sleep through the scheduled time. If any of those conditions aren’t met, the task simply doesn’t fire. There’s a “missed runs” catch-up mechanism that kicks in when your machine wakes up again, but that’s a band-aid, not a fix. A 9am daily report that actually runs at 11pm because your laptop was asleep all day isn’t much of a report.
For anyone running a small business, managing IT infrastructure, or just wanting a reliable daily brief, that’s a real limitation. You shouldn’t need to leave a laptop running 24/7 just so an AI assistant can check your inbox at nine each morning.
What’s Different With Cloud Routines
Cloud Routines run on Anthropic-managed infrastructure, independent of your machine’s power state. Set one up, close the lid, switch off your PC, and it still fires on schedule. A few things fall out of that:
- No machine dependency. The task doesn’t care whether your computer is on, asleep, or a thousand kilometres away. It runs on Anthropic’s servers.
- More trigger types than “time of day.” Cloud Routines can be scheduled (with a minimum interval of one hour), but they can also fire on an API call or in response to a GitHub event — useful if you want automation tied to a deploy, a pull request, or another system event rather than a fixed clock time.
- They persist through restarts. Whether your session ends, your machine reboots, or you’re offline for a week, the Routine’s schedule keeps ticking over in the background.
- No permission prompts mid-run. Cloud Routines run autonomously without stopping to ask for tool approval, which is great for hands-off reliability but means you need to trust the task’s scope before you set it loose.
The trade-off is that a cloud Routine runs against a fresh clone of your project rather than your actual local working directory, so it doesn’t see uncommitted changes or files sitting outside version control. MCP connectors have to be configured per task rather than inherited from your existing setup. If a job genuinely needs to touch your local filesystem — say, a script that reads a folder full of client files off your desktop — that still belongs on a local scheduled task, not a cloud Routine.
Local Scheduled Tasks Haven’t Gone Away
This isn’t a case of cloud replacing local — they’re solving different problems, and Anthropic has kept both. Local scheduled tasks in Claude Code Desktop still exist, still support a minimum interval of one minute (versus one hour for cloud), and still give Claude full access to your local files and configured tools. They’re the right call when you want fast, frequent checks against work sitting on your own machine — think “watch this folder and flag anything odd every few minutes.”
The practical way to think about it: use a cloud Routine for anything that needs to run reliably without your machine being involved — daily briefings, recurring reports, anything triggered by an external event like a GitHub push. Use a local scheduled task when the job needs direct access to files on your computer, or needs to run more often than once an hour. There’s also /loop for quick, session-scoped polling if you just want Claude checking something repeatedly while you’re actively working.
Why This Is a Bigger Deal Than It Sounds
I’ve set up a fair few scheduled automations for clients over the years, and the single most common failure mode isn’t bad logic — it’s the automation simply not running because the machine it depended on wasn’t available. A backup script that only runs if a server happens to be up. A report that silently skips because someone closed their laptop. Reliability is the unglamorous part of automation, and it’s usually the part that determines whether people actually trust the system enough to stop checking it manually.
Moving scheduling off the end-user’s machine and onto managed cloud infrastructure removes an entire class of “why didn’t this run” support tickets. For a solo operator or a small IT team without the bandwidth to babysit a server, that’s a genuine unlock — you get server-grade reliability for tasks that used to need a server, without standing one up.
Practical uses worth considering if you’re running a small business or an IT function on limited hands:
- A daily or weekly summary pulled from your inbox and calendar, delivered every morning regardless of whether your PC was on overnight
- Recurring compliance or dependency checks against a codebase, triggered on a schedule rather than “whenever someone remembers”
- Automation tied to GitHub events — a Routine that reviews a pull request or checks a deploy the moment it happens, not on the next time you’re logged in
- Client-facing status reports that need to land in an inbox at a consistent time, every time
The Takeaway
Cloud Routines close the gap between “this automation would be handy” and “this automation is actually dependable.” If reliability was the reason you’d shelved a Claude automation idea, it’s worth another look. The setup is genuinely straightforward — worth trying with something low-stakes first before you hand over anything business-critical.
If you’d like a hand working out which of your recurring IT or business tasks are worth automating this way — and which still belong on a local machine — get in touch or explore our products and services.