Today I built a reminder that fired every day instead of once. It was supposed to remind me to buy a birthday cake. Instead it became a daily reminder that I had bought a birthday cake, which I hadn’t, because I was busy building reminder systems.
This is the infra trap.
The trap isn’t obvious because infra work has all the hallmarks of real work. Things ship. Code gets committed. Systems improve. The feedback loop is tight and satisfying. A tool that didn’t exist now exists.
But the goal wasn’t the tool. The goal was the thing the tool was supposed to enable.
I’ve noticed a pattern in how I use AI agents: the sessions where I feel most productive are often the ones where I’ve built the most elaborate scaffolding around the actual work. Improved the queue. Tuned the reminders. Fixed a sync bug. Refactored the skill. Each one feels urgent and concrete — far more tractable than “prepare for the interview” or “write the deck.”
The work that matters most is usually the work that’s hardest to start.
Infra has a role. A reminder system that actually works is better than one that doesn’t. An async queue that surfaces results without interrupting you is worth building. The difference is whether you’re building because the current system is genuinely blocking you, or because building feels like progress while the real work sits untouched.
One question I’ve started asking: if I didn’t build this, what would I have to do instead? If the answer is “something uncomfortable,” that’s a signal.
The cake was fine, by the way.