The monitoring you set up before launch, not after
After an incident you will wish you had three specific things. All of them take an afternoon to add beforehand and are painful to add during.
CI that gates releases instead of just running tests
A pipeline that reports failures nobody blocks on is theatre. What we make mandatory, and what we deliberately let through.
What our AWS bill taught us about architecture
The invoice is a design document nobody meant to write. Reading ours line by line changed decisions we thought were settled.
Migrating a live database without downtime
The pattern is always the same: expand, migrate, contract. Where teams get hurt is doing all three in one deploy.