There’s a simple yet massive productivity unlock with your favorite AI agent that I’m surprised more people aren’t already using:

Run it in the cloud.

For most of the last couple of years, the default has been to have agents do their work locally. That even helped fuel the Mac Mini craze earlier this year. But increasingly, it makes sense to move this work to the cloud for many of the same reasons cloud computing was transformative in the first place.

Your agent keeps working when you close your laptop. You can start something at your desk, walk away, and come back when it’s done. With many products, you can check in or continue the work from your phone. And perhaps most importantly, you can have a bunch of agents working on different things in parallel without tying up your computer.

It’s also a cleaner security model. Instead of giving an increasingly capable agent access to your personal computer, it can operate inside an isolated cloud environment with access only to the things you’ve chosen to give it.

And setting this up is surprisingly simple.

Step 1: Put the context it needs somewhere it can access.

For coding, that’s easy: connect your GitHub repo. But this doesn’t have to be code. I’ve actually put a bunch of our business context into a private GitHub repo so my agents can work with it. For other work, you can connect things like Google Drive and other cloud services.

Step 2: Run the agent in the cloud instead of locally.

That’s it.

Oddly, this still isn’t the default experience in many AI products. In Claude and ChatGPT/Codex, for example, cloud execution exists alongside local execution, but you have to choose the right mode. Once you do, the work happens in an isolated cloud environment rather than on your computer.

There is one important caveat that applies whether you’re doing this locally or in the cloud: think carefully about where you’re storing your data and what you’re giving these systems access to. Understand your provider’s training and retention policies, particularly for sensitive data. If you’re dealing with regulated data such as PHI, make sure the specific product, configuration and data flow are covered by the appropriate controls. Many AI providers offer BAAs and Zero Data Retention (ZDR) for certain products and configurations. Make sure you’re actually using one that’s covered.

Why hasn’t cloud always been the default? Running a secure, isolated computer for every user and every agent session takes real infrastructure and costs money. Local execution was an easier place to start.

But the economics and infrastructure are catching up, and cloud execution is becoming a core part of how these agents work.

I do find something amusing about all of this. Some of the biggest AI productivity unlocks are old ideas coming back around: cloud computing, virtual machines, cron jobs, repositories.

Everything old is new again.