the things I want from "siri AI" that I'm not going to get:
- "here's how I pronounce this" so I can launch things like "Authy" with a voice command (try it)
- disable _all_ cloud interaction, even if it makes some things less reliable
- a reference sheet of commands so I can understand its capabilities
- explicit preference for "respond only to my voice"
- error handling in Shortcutsnone of these things involve an LLM, and attempting to cram LLMs into it makes most of them worse