Conventional unit tests remain essential around an LLM application, but they do not answer the most important production question about whether a change preserved useful behavior. A parser can still return valid JSON while the answer becomes less grounded. A tool call can satisfy its schema while selecting the wrong tool.