I think that at least part of my dislike of chatbots is that they force you to navigate an often poorly thought out decision tree in order to try and accomplish

I think that at least part of my dislike of chatbots is that they force you to navigate an often poorly thought out decision tree in order to try and accomplish something. When you talk to a human they're often navigating the same decision tree but they're acting as an interpretation layer to translate your natural language into the specific series of commands the machine they're working with will accept. It seems likely to me that an LLM would be able to provide that functionality.