User experience design rests on an assumption we rarely say out loud: that a person is looking at the screen, deciding, and acting. For two decades that held. A designer could think about attention, memory and emotion, and be reasonably sure those things mattered, because the thing on the other side of the interface had all three.
That assumption is quietly breaking. The "user" of a tool is increasingly not a person at all. It is a model that has been handed the tool, told the goal, and left to figure out the buttons. When that happens, the interface stops being something a human reads and becomes something an agent operates. And almost nothing we learned about designing for people transfers cleanly.
The user just changed
Think about the last time an AI assistant filled in a form for you, navigated a website, or clicked through a settings panel. The assistant did not experience the interface. It parsed it. It read the labels and the button names and the underlying structure, inferred what each control did, and acted. Its "experience" of your carefully designed screen was a sequence of tokens and a decision about which element to invoke.
So here is the question this essay is really about: if the primary operator of our software becomes an AI, are we still designing for users? Or have we quietly moved to designing for machines, and just not admitted it yet?
UX was built for a person
User experience design optimises for people. We care about visual hierarchy because eyes scan. We care about feedback because people get anxious when nothing happens. We care about delight because people form attachments to the things they use. All of it is downstream of a human nervous system meeting a screen.
An AI agent has none of that. It does not scan, it does not get anxious, it does not feel delight. What it has is a limited budget of context, a set of actions it can take, and a hard need for those actions to be unambiguous. Where a human benefits from a well-placed empty state or a clever animation, an agent benefits from a clean, predictable, legible contract: what is this thing, what can I do with it, and what happens when I do.
AI-experience design, not UX
I think we need a name for this, because pretending it is the same job lets us keep making the wrong trade-offs. Call it AI-experience design. AiX, if you want something that does not roll off the tongue. It is the discipline of designing an interface so a machine can operate it well, the way UX designed interfaces so a person could.
What does AiX optimise for? Legibility over hierarchy. Unambiguous affordances over discoverable ones. Predictable state over delightful state. The things an agent needs in order not to make mistakes. A human who misreads a button gets confused and recovers. An agent that misreads a button quietly does the wrong thing, at scale, over and over, without ever noticing.
A human who misreads a button recovers. An agent that misreads one just does the wrong thing, at scale.
What an agent actually needs
None of this is theoretical. The properties that make a tool agent-friendly are concrete. Semantic structure, so the model can tell a heading from a label from body text. Honest naming, so a button called "save" saves and a toggle called "enabled" means enabled. Explicit state, so the model can tell what already happened and what is possible next. And a thin, consistent set of verbs, so there is one clear way to do each thing.
Do not abandon the human
Here is where I will push back on my own argument. The tempting conclusion is "our users are all AI now, so design for the machine and stop worrying about people." That would be a mistake, for two reasons.
First, humans are not going anywhere. The person is still there, watching, approving, intervening. The interface has to serve both, often in the same session. Second, and more interesting: a lot of what makes an interface good for an agent also makes it good for a person. Clarity, consistency, predictability, honest labels. These were good design principles before anyone put a model in front of a screen. AiX is less a replacement for UX than a new lens on it.
The real question
So: are we designing for AI use? The honest answer is yes, and we have been doing it badly because we have been doing it accidentally. The tools that win over the next few years will not be the ones that abandon human-centred design, and they will not be the ones that ignore agents. They will be the ones that treat the machine as a legitimate user, design for its needs on purpose, and still remember there is a person somewhere behind it.
That is the design job now. Two users, one interface. If that sounds harder than it used to be, it is because it is.