i wonder how much of the disillusionment comes from the interaction mode rather than the model itself. i have a coaching background and accidentally used open

i wonder how much of the disillusionment comes from the interaction mode rather than the model itself. i have a coaching background and accidentally used open questions instead of instructions when working with claude last week. stuff like "what would break here?" instead of "check for errors." the outputs were qualitatively different. not just better, different in kind. when you only give instructions you only see execution. change the conversational style and you get something else entirely. most people (myself included until recently) talk to LLMs like they're command line tools. maybe that's why they feel like dumb text generators. not saying they're secretly brilliant. just that we might be measuring their ceiling through a very narrow interaction pattern.