I must be missing something here. I used to work on a voice assistant product and everything demoed is pretty bog standard stuff: play music, shop, order food, get Uber rides, ask random queries, recognize images.

This also runs into the same core issue as voice assistants, which is that it's way easier to transmit complex thoughts/actions on a nice big visual UI than it is to convert those thoughts/actions into verbal instructions.

For example, when people order food from an online menu, most people are not fine with "get me the closest pizza available". They're usually considering complex tradeoffs between lots of factors: alternative restaurants nearby, how visually appealing is pictures of the food, health content, other options on the menu including their prices and content and how well they complement other dishes, etc. Figuring out how to express these preferences via a back-and-forth with an AI is more tedious than just looking at a visual interface and letting your subconscious process all this information into a choice.