I never understood why they haven't had a team implementing simple features like that for over a decade. It's not even AI. Have a dashboard of the most commonly tried things every day, and develop the features.
Yes, before they went "machine learning" all these assistants were much closer to command lines you spoke to - you had to learn the phrases, but they'd work reliably once you did.
I'd love something like this to run at home. Like an open weight model with a configurable command grammar that returns the most likely thing I said that matches the grammar. I've found things that are close but they have bits missing (like support for English) or aren't really open.
