The Cost of Always Being On-Call

I built an agent last time in this series to watch my servers so I don't have to. It restarts things on its own, retries failed jobs, and writes a short note when it fixes something without me. It works. Pages that used to wake me up mostly don't anymore.

What surprised me is how little that changed underneath. The hum has nothing to do with alerts firing. Some strip of my attention stays reserved no matter what I'm doing, behind dinner, behind a walk, behind a conversation that has nothing to do with any server anywhere. Nothing has to break for that reservation to cost something. Holding it open is the cost by itself.

I run two side projects, NOGnet and Magatama, alongside the regular server fleet, and the shape of the problem repeats itself in each one. I'm the only person who knows how any of it fits together end to end. I've written documentation for parts of it. Documentation doesn't capture a system's moods, the quiet way it acts before it fails loudly, and that knowledge stays wherever I put it, mostly in my head. It doesn't clock out when I do.

Vigilance doesn't clock in

People mix this up with busyness. I used to as well. Busyness leaves a trail: commits pushed, hours logged somewhere. Vigilance leaves nothing you can point to. There's no line item for capacity held in reserve against a failure that never happens. It shows up instead as the reflex to check a screen that hasn't buzzed, or as a strange flatness in a conversation, half of me across the table and half of me still listening for something else. Nothing needs to fail for that mechanism to run. Running it is the whole cost, whether or not it's ever called on.

The automation helped with the busyness side of things. Fewer nights losing sleep over something a script could have caught. Fewer afternoons burned on a fix I could have written months earlier. I'm glad I built it, and I'd build more of it. What it didn't touch is the feeling underneath, tied less to how often anything actually broke than to being the last name on the list, the final backstop once the automation runs out of ideas. A system that heals itself ninety-nine times out of a hundred still leaves you the hundredth, and you never know in advance which hundred you're in.

So the agent hasn't relieved me of the role. It changed what the role feels like on a quiet day: less firefighting, more standby. Standby is harder to name as work, which makes it harder to justify resting from. Nobody asks how the standby went.

I don't think more automation solves this, though I'll probably keep writing it anyway. The more honest move is smaller: noticing when the reserved attention is running and treating that as a real cost instead of a mood or a personality trait. Some evenings I catch myself mid-check, phone already in my hand with nothing on the screen worth seeing, and I try to put it down without opening it. Putting the phone down unopened doesn't fix anything by itself. Doing it again tomorrow night is the closest thing I have to a plan.