I think ironically your criticism of voice interfaces actually betrays perhaps their biggest weakness over visual interfaces - it's not immediately obvious what

I think ironically your criticism of voice interfaces actually betrays perhaps their biggest weakness over visual interfaces - it's not immediately obvious what their capabilities are, what they will respond to or understand, or what your options in a situation are. Super users and developers of these systems are likely able to harness them for much more powerful usecases than the average person, but how do you make those options and usecases visible to the end user? It would seem like the answer is either 1) a robocall / callcenter bot style solution (press 1 to ____, press 2 to ____, press 3 to ____) where you must listen to lists of things but can more expressively verbally respond rather than press 1, or as you said solve the Hard AI problem of the bots contextually understanding your current situation and suggesting options based on it.