An AI assistant can give different answers for the same inputs. An application written by an AI assistant, if it doesn't just call an LLM at runtime, will mostly likely produce the same answer for the same inputs. That is, unless there's something intrinsic to its business logic that makes it nondeterministic.
Sometimes you really want the latter even if the former is less effort.
Sometimes you really want the latter even if the former is less effort.