The Interface Is Disappearing. The Human Isn't.

A few weeks ago, I was sitting at my desk trying to book a flight for a conference. I have four monitors on my desk, which on most days feels incredibly useful, but on this particular day, all four seemed to remind me how many things I still hadn't finished.

One screen had the presentation I was supposed to be working on, another had email and Teams messages coming in, my calendar was open somewhere else because I was trying to figure out which day I could actually leave, and on the screen directly in front of me were way too many tabs with flights, hotels, and conference information.

I was already in the middle of several deadlines and didn't really have time to plan a trip, but I needed to get the flight booked before prices went up. I kept telling myself to just pick one, book it, and get back to work. Of course, it wasn't that simple. One flight was cheaper but had a long layover. At the same time, I checked hotel dates against the conference schedule and made sure I wasn't solving one problem while accidentally creating another. None of this information was particularly difficult to find. In fact, that was almost the problem. I could compare flights myself, ask AI to sort through the options, and read explanations about why one option might be better than another. I had more than enough information to make a good decision, yet I sat there, surrounded by four monitors full of it, growing increasingly frustrated by a decision that should have been simple. At some point, I realized I wasn't really struggling to find information. I was struggling to understand what mattered. What happens when the work disappears?

That thought stayed with me because, right now, technology can help me make that decision. But we are quickly moving toward a world where I may not have to make that decision at all. We can already see the direction this is heading. We moved from navigating screens to telling technology what we want in conversation, and the next step is technology that simply acts in the background without being asked. Each step removes more of the interface, and with it, more of the work.

Take that same trip and move it a little further into the future. My meeting moves from Tuesday to Wednesday, but instead of opening an airline app and searching through confirmation emails, a background service notices the calendar change. It changes my flight, extends my hotel stay, and adjusts my transportation, and by the time I look at my itinerary, everything has already been taken care of.

After the afternoon I just described, that sounds wonderful. Then I notice that my new flight home has a six-hour layover. At first, I wonder why the system would choose that flight. Then I see the price and realize that it saved me $120. Maybe it has learned from my previous habits that I usually care about staying within budget. From the system's perspective, this might be a completely reasonable decision.

Nothing necessarily failed here. The AI didn't invent a flight, book the wrong day, or misunderstand my calendar. It did exactly what its understanding of my preferences told it to do. The problem is that after several days away at a conference, exhausted and ready to get home, I would happily spend another $120 to avoid sitting in an airport for six hours.

It didn't necessarily make the wrong decision. It made a trade-off on my behalf, and I didn't know it had made one.

That is the part I can't stop thinking about. The work can disappear. The need to understand doesn't.

Lately, I have been hearing a lot about the role humans will continue to play as AI becomes more capable. We talk about people evaluating and validating what AI produces, bringing context and empathy to decisions, and applying judgment.

I agree.

But thinking about a future where technology makes decisions on my behalf, largely out of sight, makes me wonder whether we're starting that conversation one step too late. How can I evaluate a decision if I don't understand what shaped it?

In my travel example, I don't need the AI to show me every flight it considered or give me a transcript of everything it did. I certainly don't want it to handle the booking and then ask me to reconstruct the entire decision through a three-page explanation. What I need to understand is much smaller: it chose price over convenience.

That gives me something specific to judge. I can recognize the trade-off, decide whether it makes sense for this trip, and make clear that getting home sooner matters more to me this time.

That isn't simply about transparency. It is about understanding, and I think the distinction matters.

More explanation isn't necessarily the answer.

When people don't understand technology, one of our instincts is to explain more, and AI makes that easier than ever. Instead of quietly booking my flight, the system could present me with five alternatives, each with departure times, fare rules, baggage policies, and a thoughtful explanation of why it might be right for me.

At that point, nothing is hidden from me, but I am also right back where I started. I scan options, scroll back up because I have forgotten which flight had the better arrival time, and try to remember which difference actually mattered to me. It has explained more, but I don't necessarily understand better.

That distinction keeps pulling me back to something very basic about being human. Our attention and working memory are limited. We filter enormous amounts of information because we have to, and we create mental models and shortcuts that make complicated things easier to understand.

AI may become exponentially more capable, but the human brain on the other side of it isn't receiving the same upgrade. The more AI can produce, process, and act on, the more those human limitations matter. And perhaps that is where the next UX problem begins. We've been here before, just not quite like this.

Mica Endsley's research into situation awareness is particularly interesting to me because it describes awareness through three levels:

Perceiving what is happening

Comprehending what that information means

Projecting what is likely to happen next. The research came from a very different technological world, but I keep seeing those same human needs in the systems we are building now.

If an AI system works quietly in the background, I still need a way to sense when something meaningful has changed. Once I notice it, I need enough context to understand what that change means to me. Over time, I also need a reasonably accurate expectation of how the system will likely behave next. That last part matters more than it sounds, because we don't do anticipation occasionally. It's how our brains work.

"Brains, it has recently been argued, are essentially prediction machines." - Andy Clark, Whatever next? Predictive brains, situated agents, and the future of cognitive science (2013)

We don't just take in the world around us; we are constantly predicting it, and we tend to notice when one of those predictions breaks. A system working out of sight doesn't escape that. It becomes one more thing we learn to predict.

Think about my travel service again. For months, it might make decisions that feel completely right. Every flight works, every hotel makes sense, and every adjustment fits what I would have chosen myself. Without even realizing it, I begin forming a mental model that this thing knows me and understands what matters to me.

Then comes the six-hour layover, and I suddenly discover that what I thought the system understood and what it actually prioritized aren't quite the same.

The interface used to give us clues.

Traditional interfaces are far from perfect, but they constantly communicate information about the systems behind them. A button tells me something can happen, a progress bar tells me something is happening, a confirmation tells me something happened, and warnings tell me something deserves my attention. Sometimes we design those clues well, and sometimes we don't, but they help us build mental models of how a system works.

Now imagine that much of that interface begins to disappear. Instead of navigating through a system, I simply express an intention: move my trip, find the best option, take care of this. 

The technology handles everything in between.

That is incredibly powerful, because so much of what we traditionally called interaction was actually work. But when we remove interaction, we may also remove some of the understanding that came with it. Even when the technology becomes invisible, the human brain still needs enough information to make sense of what is happening around it. This is where UX gets really interesting.

People talk a lot about what happens to designers when products no longer require traditional interfaces. I don't think the disappearance of the interface means the disappearance of UX. If anything, I think it forces us to remember what UX was supposed to be about in the first place.

We didn't study people because people needed buttons. We studied people because we needed to understand how they perceive information, how they make decisions, how they build mental models, how attention works, and how memory fails. The screen was simply one place where we applied that understanding.

What if the AI isn't wrong?

An obvious objection follows from everything I've said so far. If AI becomes reliable enough to work without an interface, why does the human need to understand anything at all?

Much of our current conversation about human oversight starts with the possibility that AI might fail. It might hallucinate, misunderstand a request, or produce something biased or incorrect. Those are real problems, but they aren't the problem I'm most interested in.

I'm interested in what happens when the AI works.

If it gets enough small decisions right, I will probably pay less attention to how it makes those decisions. And just like with my travel service, my mental model can quietly drift from how the system actually behaves until that difference matters one day.

A six-hour layover is annoying, but the same human problem becomes much more consequential in other environments. A financial service might make a decision because it believes preserving cash is my priority. An enterprise AI might prioritize one customer, risk, or recommendation because it has learned what matters to the organization. The technology could be operating exactly as designed and still leave the human without enough understanding to make sense of the consequence.

Even a reliable system is still making trade-offs on someone's behalf. My preferences can change from one trip to the next, and the system's picture of me can drift from what matters to me right now. That is why I don't think reliability alone solves the UX problem.

So what exactly should we design?

Maybe my travel service should simply tell me that the option it selected saves $120 but adds a six-hour layover. But the more interesting design question isn't what sentence we put on a screen. It's how we know that this particular moment deserves my attention in the first place.

I think the answer starts with the brain, not the screen. Everything I've described about how we work points to a few things we need to think about.

• Our attention is limited, so we have to design what deserves it. Not every decision needs to be seen. The ones that do are the ones with real consequence, the ones that are hard to undo, or the ones where two things I care about pull in different directions.

  
  

• We understand through mental models, so we have to design what shapes them. The smallest piece of information that changes my picture of a decision, like it chose price over convenience, is often worth more than a complete explanation.

  
  

• Our brains are constantly predicting, so we have to design for the moment a prediction breaks. If I'm going to be surprised, that surprise should help me update my understanding of how the system works, not just leave me wondering why.

These map closely to Endsley's levels of perceiving, comprehending, and projecting. And beneath them all sits a question I don't think we have answered yet: how do you study someone's understanding of a system they rarely see?

I don't think the answer is more UI, more explanation or simply putting a human "in the loop." 

Human judgment isn't a design strategy. We have to design the understanding that makes judgment possible. - Julianna G.

The interface may disappear. The human doesn't.

Regardless of how intelligent, autonomous, or reliable AI becomes, the person on the other side still has the same human brain. We will still have limited attention, build mental models, make predictions, and need to understand enough to know when to trust, when to question, and when something deserves our attention.

That is why I don't believe UX disappears with the interface. It moves. As technology becomes less visible, our job isn't to bring all of that complexity back into view. It is to understand what people need to know, when they need to know it, and how to help them make sense of what matters.

For decades, much of UX has focused on helping people understand how to use technology. As AI begins to operate more of that technology for us, I think we're facing a different question:

How do we design for human understanding when there is less and less for the human to see?

The interface may be disappearing. The human isn't.