AI doesn't replace UX — it brings in a new protagonist

You see this prediction everywhere: AI is going to be the end of experience design. If software practically writes itself, if agents carry out tasks without anyone touching a button, why design an interface that barely exists anymore?

It sounds logical, and it's wrong. It confuses UX with screens. And UX was never the screens — or at least, never only the screens.

Software exists to solve a problem

That's the real reason any software exists: there's a person with a problem and a system that has to solve it for them. AI changes the material we build with, not the reason we build. Designing the how — what the system does, when, what it shows the person, what it asks of them, what it decides on its own — that is user experience design.

What changed is where the experience lives. It used to sit almost entirely on the surface: buttons, forms, flows the person walked through step by step. Now a huge part of it has moved underneath, into processes that run on their own. But moving isn't the same as disappearing. If anything, it adds a new layer of difficulty: trust.

What you have to design when "there's no interface"

Take an AI agent that works in the background and does its job with no one watching. Someone still has to answer:

• How does the person know the agent is working, or that it already did something?
• How do they trust what it did without checking everything by hand?
• What happens when the agent isn't sure? Does it escalate, ask, or stop?
• When does it hand control back to the person, and how?
• What does the person see when something goes wrong, and how do they fix it?

None of these answer themselves just because you have a good model. They're design decisions. An agent that does everything right but leaves the person blind to what it did is a bad product — no matter how flawless the model and the harness around it are. The experience isn't in the screen that shrank; it's in those moments of trust, confirmation, and correction.

The new protagonist

UX used to design a relationship between two: person and system. Now there's a new protagonist at the table — the AI agent, which acts, decides, gets things wrong, and is part of the experience. Design stops being just "what the person sees and touches" and becomes "how the person, the system, and the agent behave with one another."

That brings in problems that didn't even exist as design questions before: how much autonomy you give the agent, when it has to ask permission, how it signals uncertainty, what tone it uses, when it should stay out of the way and when it should show itself. The agent has a way of behaving and an implicit contract with the person. Both are designed.

Experience design no longer covers just the user facing the system: it now covers the user with the agent, and the agent itself — how it "experiences" the system and the person it works with.

Where design and engineering stop being separate things

And here's the part most people skip. In an AI product, a lot of those experience decisions are, at the same time, engineering decisions. How you assemble the context the agent receives, what you retrieve and what you don't, where you place the guardrails, what fallback you trigger when the model fails — all of it defines what the person ends up living through. You can't design an agent's experience without getting into how it's built, and you can't build it well without thinking about the experience. Experience defines the architecture; architecture refines the experience.

Trust in an AI product is won or lost at that seam, between what was designed and what was implemented. A beautiful flow on top of a pipeline that hallucinates doesn't earn trust. A solid pipeline that explains nothing to the person doesn't either.

Where that leaves us

UX doesn't die with AI. It shifts, and it gets harder. Less screen, more behavior. Less "where does the button go" and more "how much can this decide on its own, and how do we tell the person about it." There's still a human problem to solve with software — only now the software has a new protagonist inside it, and that protagonist has to be designed too.

My work lives right there, where experience and engineering meet to build AI products people can actually trust.

If you're thinking through how to design the experience of an AI agent product, that's the work I do. Let's talk →