I’ve been thinking about how people build apps and what the ideal interface for it could look like.
One concept is an app builder based on sketching. You would use an iPad to draw an interface, watch it turn into a working app, and continue editing it while using it. You would only need to draw simple shapes and choose the general feeling of an element. The system would understand whether something should be a button, an input field, a list, or a navigation bar.
It sounds great in concept, especially to people like me who want to influence the design themselves.
But most people do not want to think about any of this.
When people prompt an AI to build an app, the AI adds all the things they would otherwise forget. It puts a back button where one is needed. It adds the ability to rename or delete something. It creates empty states, confirmation dialogs, settings pages, and sensible navigation.
These choices seem obvious, but they are only obvious because we have seen them thousands of times.
This is one of the reasons AI-generated apps are already starting to look and behave alike.
The AI is not inventing an interface from first principles. It is drawing from an enormous collection of existing conventions: where buttons tend to sit, what an input field looks like, how navigation works, what happens when you swipe, and which actions should be hidden inside a menu.
These conventions are useful. Without them, every app would require users to learn a new language.
But they are not neutral.
At some point, someone decided how a button should look. Someone decided that swiping from the left edge should take you back, that destructive actions should be red, and that three dots should reveal more options. Some of these patterns were designed deliberately. Others emerged gradually because enough products copied one another.
Over time, they became standards.
Now they are deeply embedded in the models building the next generation of software.
This makes me wonder how conscious companies like Apple are of the influence they have when creating a design system. A component introduced today may not only appear in thousands of apps built by developers. It may become part of the training data that teaches future AI systems what an app is supposed to be.
Design systems used to influence designers.
Now they may influence the systems that replace part of the design process itself.
That changes the importance of seemingly small decisions. The placement of a button, the behavior of a sheet, or the appearance of a selected state can propagate far beyond a single operating system. Once a convention becomes common enough, AI begins to reproduce it automatically—even when nobody explicitly asks for it.
There is something reassuring about this. Anyone can describe an idea and receive an interface that is understandable, complete, and mostly sensible.
There is also something limiting about it.
When building software no longer requires choosing every detail, fewer people will question those details. The AI will fill in the gaps with the most probable answer. And the most probable answer will usually be the thing we already know.
Prompting might therefore make it easier than ever to create new kinds of software while making it harder to create new kinds of interfaces.
The sketch-based builder still appeals to me because it keeps some of that authorship alive. It gives people a way to say: no, the button belongs here. This interaction should feel different. This object should not behave like every other object.
But perhaps that is a niche desire.
Most people do not want to design an app. They want the app to exist.
And if AI is responsible for everything between the idea and the finished product, it will not just build our interfaces.
It will set the standard for them.