I have a feeling not one of these chatbot vendors has any clue how people actually want to use computers. Most of the features available with chatbots today were also available twenty years ago; although the voice recognition is better and it has limited ability to interact about semantics the interface is just as unusable as it ever was.
This. One of the main challenges in GenAI today is UX: how do you actually embed the "intelligence" of LLMs into software in innovative ways that are not mere chatboxes, because prose sucks for showing information and using said outputs for making decisions.
Even if you make the LLM write prose in beautifully formatted markdown, you're still transfering mental complexity to the user in the way of parsing text responses.