margin notes essays on software and craft

Boring technology is a feature, not an apology

March 2026 · 9 min read

Every migration I have regretted started with the sentence “this is much more modern”. Not because the new tool was bad — usually it was fine — but because we underestimated how much of our operational knowledge was stored in the shape of the old one.

A boring stack is a stack whose failure modes you have already met. That is not a small thing. When the pager goes off at 2 a.m., the difference between “I have seen this before” and “I think this is the queue” is roughly an hour of downtime.

The real cost of novelty

Novelty is not paid at adoption. It is paid six months later, when your YAML is three versions behind and the one person who understood the upgrade path has changed jobs.

Choose new things deliberately, in one layer at a time, and write down why.

That is the whole trick. One new layer per year, with an ADR next to it, and a rollback that you have actually executed at least once on a staging machine.

On writing runbooks nobody reads

February 2026 · 5 min read

A runbook that starts with three pages of architecture diagrams is a runbook that will be opened at 2 a.m. and abandoned at 2:01. Put the command first, the why second, the diagram last.

Postgres is not a queue (but sometimes it is)

January 2026 · 12 min read

Every argument on this topic is really an argument about throughput assumptions nobody wrote down.