Engineering leads on what changed once every error came with a stack trace, a release and the user who hit it.
Our p95 on checkout was hiding behind one slow Postgres query. Tracewell pointed straight at it in the first hour.

Stack traces finally point at our code, not the bundler.

On-call pages went from about 40 a week to 12. Most were alerts nobody acted on.

The session replay sits right next to the error, so I stopped asking people to reproduce bugs.

Each regression points at the deploy that caused it. Rollbacks don’t need a meeting anymore.

Setup was one package and a DSN. The first real bug showed up before lunch.
