How to Run Claude Code Routines in the Cloud (And Why Mine All Broke in San Diego)

If you have started scheduling work inside Claude Code, one question shows up fast: how to run Claude Code routines in the cloud so they keep working when your computer is closed. Routines that live in a local folder only run while that laptop is awake, which means the system you built to work without you still waits on you being at your desk.

I ran into this the hard way. I was in San Diego over a weekend visiting a friend. I brought my laptop, and then I spent the days out with people and never opened it. My computer was shut. Every one of the routines I have running in Claude Code failed to fire, not because anything was broken, but because they need the localized version open and something running all the time. On a call with a founder I coach, I said out loud what the actual problem was: everyone operating at the top of this space is running into the same infrastructure problem, which is how there can be no difference between me being at my desk and me being in California and my ability to use Claude Code to build something that gets deployed.

What Is a Claude Code Routine?

A routine is a scheduled job. You define the steps once, you set a time, and Claude runs it on its own. I have 33 of them built right now. They handle the small repeatable work that used to sit on my plate and quietly pile up.

Here is the language shift that matters. On the call, I asked the founder whether they were using routines. They said not as much as they should. I told them the wording was telling me everything, because I do not use routines, I create routines. Routines are designed to run without me. If you are still opening the thing and pushing the button, you have a saved prompt rather than a routine.

That distinction is the whole reason cloud execution matters. Work designed to happen without you should not need you awake, at your desk, with the lid up. If you are still deciding what belongs in a routine at all, start with skills first and let the repeatable ones graduate into scheduled work.

Why Do Claude Code Routines Stop Running When Your Laptop Is Closed?

Because of where they live. When you create a routine, you choose the folder it runs against, and most people point it at a local folder on their own machine. That folder is the routine's workspace. It reads context from there and writes results back there.

So the routine has a hard dependency on one physical computer being powered on and awake. Close the lid and the schedule keeps ticking with nothing underneath it. Get on a long flight and the same thing happens.

The founder I was coaching described their own version of this without realizing it. They told me they have a laptop they travel with, a desktop they use in one room because it has three screens, and time blocked on their calendar to work from completely different environments. That is a multi device life running on single device infrastructure. It works right up until the day it matters.

How Do You Run Claude Code Routines in the Cloud?

The move is to stop treating your context vault as a folder on one machine and start treating it as something that lives online, then point the scheduled work at that.

The path I am building toward uses GitHub as the home for the context vault. You keep a local version, you keep a cloud version, and the two stay in sync. Then the routine runs against the cloud copy. You stop worrying about the integrity of the system because the system lives in the cloud rather than on whichever laptop you happened to bring.

There is a second half to this, and it is the part most people skip. Your knowledge base has to be reachable from the cloud too. My single source of truth is Notion, and that goes everywhere no matter what device I am on, which is why Notion works as an AI operating system rather than a note app. A routine running in the cloud against a knowledge base trapped on your hard drive has solved half the problem.

Watch for the prompt inside the tool as well. An hour before that call, something popped up on my screen offering to run the session in the cloud, and the exact wording was that the cloud keeps working even when your computer is off. The rails are being laid while we talk about it.

Is It Worth Running Claude Code Routines in the Cloud Right Now?

Run the honest math before you rebuild anything. My filter is whether the juice is worth the squeeze.

The cost is real. Moving everything to the cloud means reverse engineering everything you have already built and finding out what breaks and what does not. If you have three routines that fire while you are already at your desk, the answer today is probably no. Keep building and revisit it.

The case flips when the count climbs. Once you have enough of them firing daily that you have stopped tracking them by hand, and your work happens across a phone, a laptop, and a desk machine, the local dependency becomes the ceiling on the whole system. That is the point where the rebuild pays for itself.

And here is the compounding argument. We talk about infrastructure constantly because you are never going to get an award for your context vault in the moment. Three years from now you say you are thankful you built it. Same as the systems thinking behind agentic AI, the return shows up later and shows up large.

What Should You Build First, the Project or the Infrastructure?

The founder asked me this straight, and it is the best question on the call. They had a project they were excited about and a piece of infrastructure that would carry every project after it. They wanted to know which one comes first.

Infrastructure. Every time. The project is one outcome. The infrastructure is the multiplier on every outcome after it. Nobody is doing backflips about setting up a cloud based system using Notion and GitHub, and that is fine. You are putting the oil in the car, and later you are grateful for the oil.

The trap is that the project always feels more urgent, because the project is the fun part. Building at the speed of your imagination stops being possible the moment your imagination outruns the machine underneath it.

What Happens If You Have to Start Over to Run Claude Code Routines in the Cloud?

You might. I might. And that is survivable.

When the founder went looking for their own routines, they could not find them, because they had been working in the browser rather than the desktop app the whole time. Their count was effectively zero. My read for them was that zero is not a bad number right now, because I may have to go back to zero myself and rebuild everything on brand new infrastructure.

The reason zero is fine is that the thinking survives the rebuild. If your processes are written down somewhere platform neutral, moving them is a port rather than a reinvention. If they only exist as things you did once inside a chat window, you are rebuilding from memory. This is the same reason Claude Code is worth learning even without a coding background. The leverage is in how you define the work, not in which surface runs it.

Your Next Three Steps

1. Count what you actually have. Open your routines list and confirm how many are scheduled, how many fire daily, and which folder each one points at. Plenty of people find a smaller number than they expected.
2. Check for silent misses. Pick a routine that should have fired on a day your machine was closed and confirm whether it ran. A routine that quietly skipped is worse than one that failed loudly.
3. Move the context before you move the routines. Get your knowledge base somewhere reachable from anywhere. The scheduled work is easy to repoint once the context it depends on is already online.

The Standard Underneath All of This

The goal is a system where there is no difference between you at your desk and you somewhere else entirely. That standard is worth holding even while the tools catch up to it.

Routines that need your laptop open are a good first step and a bad final answer. Design for the version where the work happens whether or not you are there, then go be somewhere.