Today a product is a set of pre-drawn screens. The team tries to guess: which button to place, which filter to add, in what order to show the steps.
With AI, part of the interface will be assembled around the person’s intent. Not “find the right section” but “show me the expenses that grew and explain why”. The system will choose the data, the presentation and the next question itself.
This won’t do away with interfaces. When an action is frequent, risky or requires precision, a button and a table are more reliable than a dialogue. Nobody wants to confirm a payment by voice or look for the right row in a half-screen answer.
But static screens will stop being the only way to use a product. A good product will combine two modes: the familiar route for repeatable tasks and an AI layer for non-standard ones.
A simple thing follows from this: you need to design not only screens but also user intents, available data and the boundaries of actions. Otherwise the AI interface will become a polite but useless search operator.