AI Engineer

This is the infrastructure category — the tools you need after the prototype works and before anyone else depends on it. Evaluation harnesses, tracing, prompt versioning, guardrails, deployment, cost monitoring. None of it is exciting and all of it decides whether an agent survives contact with real users. If you are looking here early, the most valuable thing to add first is usually evaluation: without it, every change to a prompt is a guess, and nobody can tell you whether last week was better.