Chat won't replace your apps.

There are two big claims about where computing is headed. One says we will just talk to our computers and throw the buttons away. The other says the buttons are sacred and chat is a fad. The screenless AI gadgets, the lapel pin and the little orange toy, bet everything on the first claim and flopped. Not because the AI was bad, but because talking is a slow way to turn on a light. Both claims are wrong, and they are wrong in the same way.

Buttons win the simple stuff, and always will. You drag a file to the trash. You pinch a photo to zoom. The thing you want is right there and you act on it, with no sentence to be misread. For what you do every day, the coffee order, the ride home, the lamp by the door, a tap your thumb already knows beats a paragraph every time. Wedging a chatbot in there solves a problem nobody has.

But buttons have a limit. They work while the choices fit on the screen. Past that, the screen fills with menus inside menus, and every serious tool ends up looking like a cockpit built for nobody. Open a professional video editor and you see it: hundreds of controls, almost none of them yours. This is where chat sounds smart again, and where it also fails, because a wall of text is not a control panel either. Both ideas break in the same place: the huge middle, where most real work lives.

The real question was never chat versus buttons. It is how much the machine should guess. A button guesses nothing and makes you do everything. A blank chat box guesses everything and makes you describe what you would rather just adjust. The answer sits between them, and it is finally possible: the machine reads your request, decides a control beats a paragraph, and builds the two sliders and one map your question actually needs. Then it throws them away. Designers stop drawing fixed screens and start designing the parts the machine assembles on the spot. It is the biggest shift in interfaces since the desktop was invented at Xerox.

So the future is not one winner. It is all three, matched to the job. A button when the task is one tap you do daily. A built-on-demand interface when the task is specific and would otherwise drown in menus. An agent when the work is big and boring and you would rather not watch. The new skill is routing: knowing which of the three a moment wants, and refusing the other two. That is where we spend our attention now, building fewer finished screens and more kits the machine can draw from. The empty chat box was always a placeholder. The job is to fill it back in, on demand, with exactly the controls the moment needs and nothing it does not.