Your AI Routines Should Run in the Cloud

A scheduled AI routine that lives on your laptop only runs while your laptop is open. Close the lid, and your 4pm timesheet check silently doesn’t happen. Move that same routine to the cloud and it runs on Anthropic’s servers, on schedule, every time, whether your machine is on a desk, in a ute, or switched off entirely. If you only take one thing from this post, take that.

I picked this up again this week in a session with one of our Future Builder members. He had done the hard part: a daily routine that checks timesheets at 4pm, flags who hasn’t submitted, and posts the summary into his team’s group chat. Genuinely good automation. The catch was that it was running locally, which means it only worked when his laptop happened to be open at 4pm. The fix took minutes: recreate the routine as a cloud schedule and it fires every day regardless.

What an AI routine actually is

A routine is a scheduled instruction to an AI that runs without you. You write the instruction once, set the schedule, and it does the work and reports back. No dashboards to check, no reminders to yourself. The routine checks, and it tells you.

The ones I run and recommend for a building business:

1. Morning briefing, 6am. Pulls your emails, messages and calendar into one summary before you’re on site. You start the day knowing what actually needs you.
2. Timesheet check, 4pm daily. Flags anyone who hasn’t submitted, posts to the team chat. The member I mentioned had this catch a double-submitted timesheet within its first week. That’s real money.
3. Budget audit, weekly. The AI goes through every budget line item on every live job in Wunderbuild and flags anything running over. You don’t review every line on every job anymore. You investigate the flags.
4. End-of-week team summary. One post, whole team, what happened this week.

Notice the shape of all four: the AI does the checking, and you only spend attention where something is wrong. That’s the whole point. A dashboard still needs you to look at it. A routine taps you on the shoulder.

How to tell if your routine is running locally

If you built a scheduled task inside a desktop AI app on your own computer, it’s local. The test is simple: would this run if my laptop was shut in a drawer? If the answer is no, it’s local.

Cloud routines run on the provider’s servers. In Claude, these are the scheduled tasks you set up through the web app rather than your desktop. Same instructions, same output, different place of execution. The migration is usually just copying your routine’s instructions across and setting the same schedule.

Start with one

If you have no routines at all, don’t build four. Build the morning briefing first, because you’ll feel the value on day one and it teaches you how to write instructions the AI can follow. Then add the timesheet check, because compliance problems are visible, measurable and annoying, which makes them the perfect first automation.

And write the instructions down properly, in a document the routine reads, not buried in the routine itself. When the routine gets something wrong, you fix the document. I’ve written more about that in the post on giving your AI proper SOPs.

One warning before you automate everything in sight: automation that nags too much stops working, because your team stops reading it. I’ve written about that trap separately too, because it deserves its own post.

If you want this set up in your building business rather than working it out solo, this is exactly the kind of thing we do inside Future Builder and in my AI consulting sessions.