My AI Agents Work All Night. So Did I

I built a machine that can work all night, and somehow, without even noticing, I started doing the same.

There were nights when nothing was on fire. No deadline, nothing urgent. I was just checking on an agent, and then another one. One was blocked, so I unblocked it. Another had finished, so I gave it something new to do. The machine was doing exactly what I built it to do. The problem was, I was still there.

A simple wish

It started with a simple wish. I wanted my agents running all the time, not just when my laptop was open. The answer I found was my own VPS. Nothing fancy: four CPUs and sixteen gigs of RAM, running day and night. It didn’t need to be powerful. It just needed to stay on.

It turned into much more than a server. My coding agents live there now. I can open a tmux session from my laptop, or from my phone when I’m out, and pick up exactly where I left off. AI stopped being a chat window I opened now and then, and became part of my workspace. At first, it felt amazing.

Then it shifted

If an agent finished, I could give it another task. If one got stuck, I could unblock it. And if the server was sitting idle, I felt like I was wasting it. Because I could check everything from my phone, I did. After work, on weekends. It was always just a quick look, and that quick look kept turning into another hour.

The strange part is that it never felt like work. It felt like curiosity. But my head was still at work, long after I was supposed to be done. The agents never asked me to keep working. I just kept giving myself reasons not to stop.

Better boundaries, not less automation

A VPS doesn’t know it’s Sunday, and it doesn’t care that it’s midnight. I’m not built like that. So I’m trying to build a different kind of automation.

  1. Every job knows what done means. No endlessly finding more work for itself. And blocked doesn’t mean I fix it tonight. Sometimes it just waits until tomorrow.
  2. Time where I simply don’t check. That’s been harder than setting up any of the software. Not opening the terminal is the hard part.
  3. Only things that can safely wait. I don’t want an agent making decisions for me just because I wasn’t around.
  4. Stop chasing a hundred percent. I pay for the machine, so part of me wants to squeeze every bit out of it. But maxing it out was never the goal. An idle server is fine. It doesn’t need a break. I do.

I’m still figuring this out. I still check too often, and I still feel that pull when something’s blocked. But at least now, I notice it.

This is the other side of Don’t Automate the Fun. I want AI to handle the repetitive stuff so I can stay close to the parts I enjoy. I also don’t want to automate away the moment I stop.

If you’re running agents or automations too: how do you switch off? Not how you keep them running, but how you stop thinking about them.

Get new posts by email.

Leave a Reply

Your email address will not be published. Required fields are marked *