Why is LLM evaluation important in production AI systems?

Evaluation prevents teams from judging model quality only from a few impressive examples and helps detect regressions after prompt or model changes.