Hamel Husain says not every failure mode needs an automated evaluator

Hamel Husain advises against building automated evaluators for every failure mode discovered during AI development. He argues that teams should weigh the costs of different evaluator types, such as code-based checks versus LLM judges, before building an eval.