I think you've highlighted the real engineering challenge of using LLMs - overcoming the uncertainty. In some contexts, a reasonable level of uncertainty is fine, as the tool itself is being used by experts ( ie the coding tools), but in other cases, you need to engineer a lot more guardrails to ensure a result of sufficient quality.
Hilarious take.