Apple's history with accessibility is, on the whole, pretty good. I strongly suspect that the "coming soon" part of this means "after we integrate Google Gemini models into the system," so I don't think you should use the current state of Siri as a yardstick. (I actually have decent luck with the current Siri, but I don't push it very much and have sort of adapted myself to its limitations; on the flip side, I have a lot of skepticism around LLMs, but they're really a quantum leap in natural language processing capability over what came before, and the use cases they're showing here seem to be right in the LLM wheelhouse -- with the asterisk of "you're still always going to have to check its work.")