yeah i had the same knee-jerk question. I can see the use-cases here for non-audio. But for the most part it would seem that audio input get to a level of performance that it simply works 99% of the time with little training.

I just broke my screen a few days ago, and b/c of the lock-down here (im in a tiny city) everything was closed. So i was forced to use audio inputs as much as possible, and on the IOS i was horrified that i couldnt simply say "Siri, close this modal" or "Siri, send new whatsapp text to Jane"... shocked.