SQL query optimization agent accuracy improved 59% using LLM Observability

An AI agent autonomously ran 23 experiments to improve a query optimization agent from P=0.54 to P=0.86. LLM Observability Experiments tracked every hypothesis, result, and failure along the way.