A UI taxonomy for AI apps that work: surface early, prioritize accuracy and trust. ChatGPT-style apps often fail not because of the model, but because they mix different work modes. Stripe, Shopify, Intercom, and Spotify show what done right looks like.
The UI usually surfaces early and also focuses, has a very clear focus on accuracy, verification, trust. Some examples of these are Stripe, Shopify, Intercom, Spotify, and some others. So this is our current taxonomy or my current taxonomy that helps me understand a bit better what is there today. And then if we go to why any of these matters, that these categories, they map closely to these different modes of work. So what I've seen in weaker ChatGPT apps is that they struggle not because of the model or the UI itself, the UI widget itself, but because they mix the modes unintentionally.
