Building safety tooling for risk-free AI tuning of Postgres: Fast cars need fast brakes

Optimizing your database with AI is a tantalizing prospect, but how can we make sure to do this in a risk-free manner? In this talk, I will share my experience with building safeguards and guardrails for automated PostgreSQL tuning to help you sleep well at night — the better the safety net, the more freely we can let the agent work to improve the system.

I will walk through the tried and tested safety patterns: memory and performance monitoring, and validation techniques that ensure every change is safe. The goal is simple — get the performance gains you want while minimizing risk to your system. Whether you are considering automated tuning or building your own tools, building for safety should always be the highest priority.