Interesting thought experiment. I think accuracy is table stakes for a voice control system, I think the bigger challenge for adoption is discoverablility. It's hard to know what the exact syntax the voice prompt expects and everyone is different. I do wonder if LLMs and their ability to parse natural language will rapidly unlock that type of thing because they're a lot more flexible about how you command them