How AI Rewrites Mobile Design Rules

I spent the first decade of my career designing mobile apps that obeyed a simple contract with the user: tap a thing, get a predictable result, fast. Lists scrolled. Buttons submitted. Forms validated. The hard problems were about hierarchy and flow, not about whether the app would even do what you asked. AI quietly tore up that contract, and most of the design patterns we relied on no longer hold.

Working on MyoScore and PrettyType forced me to relearn the craft. Here is what actually changed once an unpredictable model sat in the middle of the experience.

Results are no longer deterministic, and the UI has to admit it

Classic mobile design assumes that the same input produces the same output. AI breaks this. The same photo can yield a slightly different score on two runs. A model update can shift results overnight. If your interface pretends to be precise when the engine underneath is probabilistic, you are lying to the user, and they will catch you.

We design now for confidence, not certainty. Instead of a single hard number presented like a fact carved in stone, we show ranges, we soften the language, and we make it clear when the app is estimating versus measuring. The goal is not to hide the uncertainty. It is to make the uncertainty feel honest rather than buggy.

A deterministic app that is wrong feels broken. A probabilistic app that is wrong feels human, but only if you designed it to.

Latency is now a design material, not an edge case

A traditional screen either loads instantly or it is broken. AI introduces a third state that lasts seconds, sometimes longer, and it happens on the most important interaction in the app. You cannot spinner your way out of this. A spinner says nothing is happening. During inference, a lot is happening, and the user wants to believe it.

• We narrate the work. Instead of a generic loader, the screen describes what the model is doing right now, step by step.
• We use the wait to set expectations. Telling someone the analysis takes a few seconds before they tap is worth more than any animation.
• We never let the wait feel empty. Even a well-chosen sentence about what is coming reduces the perceived time dramatically.

Empty states became the most important screen

In a normal app, the empty state is a polite placeholder you design last. In an AI app, it is the moment of maximum doubt. The user is about to hand over a photo of their face or their body and trust a machine to judge it. If that first screen feels cheap or vague, the trust never forms and they never reach the magic.

I now spend more time on the screen before the AI does anything than on the screen that shows the result. The pre-result screen has to do emotional work: explain what will happen, why it is safe, and what the user will get. It is the handshake before the conversation.

Inputs got messy, and that is the point

Old mobile design loved constrained inputs. Pickers, toggles, dropdowns, anything that prevented the user from giving you garbage. AI thrives on messy, natural input. A photo taken in bad lighting. A vague description. A half-formed question. The design challenge inverted. Instead of constraining input to protect the system, we now coach input to get the best from the model.

That means gentle guidance instead of hard rules. Show a live frame guide for the photo rather than rejecting it after the fact. Suggest, do not block. The most respectful thing you can do is help someone succeed on the first try rather than scolding them on the second.

The new craft is managing trust over time

The deepest shift is temporal. A button either works or it does not, and you judge it in a second. An AI feature earns or loses trust slowly, over many interactions. One absurd result can undo a month of good ones. So the design job stretches beyond the screen into the relationship.

We added small honesty mechanisms everywhere: a way to flag a result that feels wrong, plain language about how the score is computed, and an unmistakable line between what the app knows and what it is guessing. None of these are flashy. All of them are why people come back.

If you are a designer moving into AI products, drop the instinct to make everything feel slick and certain. The new craft is making something genuinely uncertain feel trustworthy anyway. That is harder, and far more interesting.