Adopting untested solutions almost always ends in tears.

And the output of an LLM is a nightmare to properly test, and to some extent it’s not testable at all because its results are not repeatable and there’s no explainability.