When machines pay attention, what deserves ours?
For decades, digital products competed for our attention. The web put information everywhere, social networks turned it into a constant stream, and smartphones made it always accessible, making attention the scarce resource. Entire industries were built around capturing it: feeds, notifications, recommendations, autoplay. The basic logic was simple. If information is everywhere but attention is scarce, the product that wins is the product that gets our attention.
We became very good at this.
Every app learned how to ask for another minute, another click, another reason to come back. Notifications turned events into interruptions, feeds turned waiting into scrolling, and the red badge turned an empty corner of an interface into something that felt unfinished. We designed systems around the assumption that attention was something to be acquired. AI changes the equation. Not because information is suddenly everywhere. It already was.
What has changed is the cost of having something else pay attention for us. An AI can read every email, search hundreds of pages, compare products, monitor a project, or watch for changes that no human would have the patience to track.
Machine attention is becoming cheap.
Human attention is not, and that creates a new bottleneck.
The question is no longer simply how a product captures our attention. It is what the product decides is worthy of our attention in the first place. When an agent filters the world before it reaches us, it is not just retrieving information. It is allocating attention, deciding what to surface, what to summarize, what to ignore, and what can safely disappear. That makes attention part of the system we are designing. And it gives designers a different question to answer:
What is worth bringing to a human at all?
We have already been trying to solve this problem
We have been quietly building systems for attention management for years. Consider the evolution of the smartphone notification. It started as useful infrastructure: something happened while you were somewhere else, your message arrived, your ride was outside, your calendar event was approaching. Then every app discovered that it could use the same channel.
A shopping app wants to tell you about a sale, a news app wants to tell you something is breaking, a social app wants you to know someone liked something, and a game wants you to come back. The notification became less like a message and more like a bid. Everyone was asking for the same scarce resource: a few seconds of attention.
Research has described this tradeoff for years. Notifications provide awareness, but they can also interrupt the task already occupying our attention, and research into notification
deferral has treated interruption as something that can be scheduled rather than simply delivered immediately. The interesting part is that the technology is finally catching up with the problem.
Your iPhone already has a primitive version of an attention allocation system. Focus modes let you decide which people and apps can reach you, Sleep can silence almost everything, and Apple’s Reduce Interruptions Focus uses Apple Intelligence to determine which notifications might require immediate attention and silence the others. The important idea is not the feature itself. It is the direction. We are moving from notification delivery to attention governance.
The phone is no longer simply asking “Did something happen?” It is starting to ask, “Does this deserve to interrupt you?”
That is a much more consequential question.
AI makes the filter the interface
The traditional interface exposes the world and asks the user to navigate it: open your inbox, browse the feed, search the database, check the dashboard. The user does the filtering.
AI changes that relationship. The agent can do the filtering first, reading the inbox before you do, comparing products before you do, and monitoring information while you are doing something else. This is why the rise of agents matters beyond simply adding a chatbot to a product. The system is no longer waiting for a human to navigate toward information. It can decide that some information should come toward the human.
That changes the economics of the interface. If a human can reasonably inspect ten things, the value of an agent may not come from finding ten interesting things. It may come from examining ten thousand and deciding that only three deserve to be seen.
The work happens upstream and the attention happens downstream, which means the quality of the product starts to depend on the quality of its omissions.
The economics of omission
We have spent the last twenty years optimizing the economics of attention capture. AI introduces the economics of attention omission. Imagine two agents monitoring the same world. The first sends you thirty seven updates every morning, technically doing its job because nothing was missed, while the second says there are two things you need to know. The second system may actually be doing much more work. It has searched, compared, evaluated, and discarded before asking anything from you.
That changes what “good” looks like.
A good agent does not maximize the information it gives you. It minimizes the attention it demands while preserving what matters. For years, the instinct was to add visibility through dashboards, recommendations, and notifications. The next generation of products may have to become much better at disappearance.
The best feature may be the thing you never have to see.
We have been here before
There is an interesting parallel with the AI conversation itself. In 2023, tens of thousands of people signed an open letter calling for AI labs to pause the training of systems more powerful than GPT 4 for at least six months. The letter described an industry accelerating rapidly and questioned whether increasingly capable systems should continue to be developed without stronger planning and governance. We’re experiencing the same situation in 2026 with multiple AI leaders asking to slow the AI race pace.
Whatever one thinks of that particular proposal, the underlying tension is revealing. We were building systems that could process and generate more than any individual could meaningfully keep up with, while simultaneously worrying about what would happen as those systems became even more capable. The same tension now exists inside the product itself. If AI can generate ten thousand possibilities, someone still has to decide which one matters. If it can read ten thousand documents, someone has to decide which three deserve attention. If it can monitor everything, someone has to decide what constitutes an interruption. Scale does not eliminate judgment. It moves judgment closer to the point where something becomes visible.
This is also why designers feel strangely threatened
There is an obvious anxiety in design right now: if AI can generate an interface, prototype a flow, write the copy, produce the illustration, and ship the code, what exactly is left for the designer to do? That question is usually framed around execution. It is the wrong level of abstraction. The more interesting problem is that generating something that looks designed has become cheap, so many more people can participate in design. A founder can prompt a prototype, an engineer can generate a UI, and a product manager can ask an agent to redesign a flow.
The designer’s problem is no longer simply that AI can do parts of the job. It is that the number of people making design decisions is expanding rapidly. That makes judgment more important, not less. Who decides which generated ideas are acceptable, which interactions are manipulative, which notifications deserve to exist, and which things should remain invisible? These are not questions that a component library can answer. They are questions about what the organization believes a good experience should be.
Nielsen Norman Group recently described this emerging situation as a “custodial era” of UX: AI makes it possible to generate content, prototypes, and working features much faster, while UX increasingly has to evaluate, triage, and govern what gets produced. Their recommendation is to encode UX knowledge into the systems generating the work through design systems, interaction patterns, accessibility requirements, and explicit rules about what should be avoided. That is not design disappearing. It is design moving upstream.
From design systems to attention systems
We have spent years building design systems so organizations can make interfaces consistently. The next layer may be an attention system. A design system tells an organization how a button behaves. An attention system tells an organization when a button, notification, message, or agent is allowed to demand something from a person.
It might define rules such as: a notification should not interrupt an active task unless its urgency exceeds the cost of interruption, an agent should summarize before escalating, a recommendation should explain why it is being surfaced, and a system should distinguish between something that is new and something that is important. An agent should also be able to say: “Nothing requires your attention.” The last one might be the hardest.
Most software has historically had a business reason to keep us engaged. The attention economy trained products to see an empty state as a missed opportunity. AI creates the possibility that an empty state is actually the product doing its job.
The agent becomes the new intermediary
This is where the economics become more interesting. For most of the internet, there was a relatively direct relationship between the product and the user. Then platforms became intermediaries, deciding what content we saw, which posts were ranked, and which recommendations appeared. Now agents introduce another layer. The agent may sit between the person and almost everything else, choosing which news stories to summarize, which emails to surface, which products to compare, and which messages can wait. The agent becomes a personal information intermediary. That creates a new incentive problem.
If the agent is optimized around the user’s goals, its job might be to protect attention. If it is optimized around engagement, advertising, transactions, or retention, it has very different incentives. One system might say:
“You don’t need to see this.”
Another might say,
“You might also be interested in…”
Both can feel helpful, but they are making fundamentally different claims about the value of your attention.
The designer’s job becomes deciding what deserves attention
This is where the role of the designer changes most. The old question was:
How do we get someone’s attention? Then it became: How do we make that attention useful? Now it becomes:
What deserves attention in the first place?
That sounds like a small change in wording. It is not. It moves design from presentation toward judgment.
Designers have always made decisions about hierarchy: what is large, what comes first, what is hidden, and what is emphasized. AI extends that hierarchy beyond the screen. Now the system can decide whether something should appear at all. That means designers have to think about thresholds, interruptions, confidence, and timing. They have to decide not just how something should look when it reaches a person, but whether it should reach them at all. This is why the emerging idea of AI agents as users is also important.
Nielsen Norman Group argues that agents are beginning to interact directly with interfaces designed for humans, forcing designers to reconsider what “the user” even means. But there is another user in the system that is easy to forget. The human who never sees the interface because the agent already handled it.
Attention is becoming part of the architecture
Attention used to be treated as a psychological outcome of interface design. Now it is an architectural decision. A system has to decide what gets processed by the machine, what gets processed by the human, and what never needs to cross the boundary between the two. That boundary is becoming one of the most important parts of product design.
Consider sleep. The reason Sleep Focus works is not because the world stops producing information when you go to bed. Your messages still arrive, your calendar still changes, and your apps still have something to tell you. The system simply agrees that most of those things can wait. That is a powerful design decision.
The world continues and the machine continues, but you don’t have to.
AI gives us the possibility of making that decision dynamically instead of forcing the human to manually configure every exception. But that also gives the system authority. If an agent decides what is worth waking you for, it is making a judgment about the value of your attention at a very intimate level. The question is no longer just whether the system is accurate. It is whether we trust its priorities.
The scarce resource is permission
Maybe attention itself is not the most useful way to think about this. The scarcer resource is permission to consume attention. Every notification asks for it, every agent escalation asks for it, every recommendation asks for it, and every interface that appears in front of you makes a claim that whatever is behind it is worth the interruption. AI makes those claims cheap to produce.
If everyone can generate content, recommendations, messages, and notifications, the amount of material competing for human attention will keep growing. The answer cannot be to become better at looking at all of it. We cannot scale human attention to match machine output. The answer has to be better filtering, better judgment, and better governance, with better systems for deciding what deserves to cross the boundary. That is the real shift.
The scarce resource is not information. It is not even attention. It is permission.
Designing for restraint
The attention economy taught us how to make products that people could not stop looking at. The AI economy may force us to build products that know when to stop talking. That is a very different kind of craft. It asks designers to care about silence as much as visibility, omission as much as inclusion, and interruption as much as interaction. It asks a product to understand that “nothing” can be the correct answer. And perhaps that is the strange promise of AI.
The machine can look at everything so that the human does not have to. But that only works if we give the machine a specific responsibility: not to bring everything back to us, but to decide what is worth bringing back. Execution is easier, generation is effortless, information is everywhere, and even machine attention is cheap. What remains scarce is the human ability to care about something, think about it, decide about it, and give it our time. So the question for designers is no longer simply:
How do we get attention?
It is:
What is worth asking a human to pay attention to at all?
From AI × Taste
A monthly series exploring the economics of taste in the age of AI:
- AI made everyone a creator, not a designer
How AI made generation abundant, making design judgment and taste the new competitive advantage. - The organizational costs of low taste
How weak design judgment compounds across products, teams and organizations. - Taste cannot be delegated
Why AI can generate endless possibilities, but deciding what should exist remains fundamentally human. - Designing for the Proxy
Why AI is changing who reads first, and what happens when we optimize for the intermediary instead of the person. - When the canvas starts acting, who’s really in control?
How interfaces are changing what designers control, from interactions to agency, orchestration and attention. - The economics of attention in the age of AI. When machines pay attention, what deserves ours?
The economics of attention in the age of AI was originally published in UX Collective on Medium, where people are continuing the conversation by highlighting and responding to this story.
