It was fully-baked in the sense that it was novel and could recognize fairly basic voice commands including simple questions of fact. Once the novelty wore off, a lot of people concluded that wasn't actually that useful given they had a computer in their pocket anyway.

Voice is still just an implementation detail. I'd argue that we still aren't at the point where general purpose computer assistants are all that useful relative to humans. It's the same 10% (or whatever) problem as self-driving has.