OpenAI has released GPT-6 to everyone, and the most consequential thing about it is not the model. It is the shape of the answer. The new “intelligent UI” renders responses as checklists, cards and interactive layouts instead of the plain stream of text that defined the first chatbot era. The chat box, after a decade of imitation, is being retired by its inventor.
The shift deserves more scrutiny than a launch-day cheer. A paragraph makes an argument in sequence and lets you watch it think; a card makes a claim and lets you tap it. When an AI answers in components, it decides not only what to say but what kind of object the answer is (a plan, a comparison, a form), and every act of formatting is an act of editing. Users noticed. Reaction among developers and power users has been split, with some calling the heavily formatted output clarifying and others calling it condescending, software that treats you like a child. Both reviews describe the same fact: the machine now has opinions about presentation, and presentation is persuasion.
There is also a business model hiding in the layout. OpenAI has been introducing advertising formats in and around its products, and cards and modules are, not coincidentally, the native geometry of sponsored content. None of that makes the interface bad; search engines walked this exact path from ten blue links to answer boxes to the ad load of today. It does mean the UI should be reviewed like an editorial product, not a neutral window: what gets a card, what gets buried in text, what gets a button.
The system card adds the necessary sobriety. OpenAI’s own documentation acknowledges statistically significant regressions on some safety evaluations, including self-harm and extremism measures, versus its predecessor GPT-5.6. Capability and safety do not automatically rise together, and a more engaging interface on top of a slightly less safe model is a trade the industry keeps making quietly.
Review the release as two products, then. Product one is a stronger model whose numbers will be argued over by benchmarks for a month. Product two is the decision that AI answers should look like apps, a decision that will outlast the model itself. The chatbot era asked: can it talk? The interface era asks a harder question, in cards: when it stops merely talking, whose interests is the layout serving?
