The annoying things is that all the use cases you just described could be handled locally (over wifi/home-nework) without Alexa being connected to the wider web.
This is all I want; give me the option to not have the assistant connect to the web but still work in a more limited capacity for home automation and to interact with the local files/programs on the computers in my home.
I was under the impression that aside from a limited realm like the wake word, it's Hard to do voice recognition without centralized processing. Some combination of faster CPU (even with network latency detracting), whatever storage requirements there are for samples or fingerprints of every possible phoneme, 0-day neologisms (okay, strictly home automation wouldn't need that part), etc. etc. etc.
This is all I want; give me the option to not have the assistant connect to the web but still work in a more limited capacity for home automation and to interact with the local files/programs on the computers in my home.