I will keep reviewing and adjusting any AI generated code until the models can consistently one-shot produce something I would consider acceptable.
I haven't been able to just "trust the model".
But anyway, trust is earned, applies to humans, companies, products, AI.