If the prompt is good, the LLM will tell you when it's wrong, but you can use production testing if necessary like Tesla.