AI does not end the attention economy. It compresses it into the moments where judgment, permission, and accountability live.
In 1971, the economist Herbert Simon saw the next fifty years coming. Information was about to become abundant, he argued, and every abundance consumes something. What information consumes is attention. His conclusion became the most quietly influential sentence in technology: “a wealth of information creates a poverty of attention.”
It took three decades and a few billion smartphones to cash that sentence in. The business model that won the internet treated attention as territory to occupy. Feeds, autoplay, infinite scroll, the red badge. Success was measured in occupied time: session length, daily actives, minutes per visit. Engagement was not a side effect of the design. It was the product.
AI looks like more of the same. More content, more synthetic noise, more things shouting into the same poor attention. That reading misses the structural change.
A feed needs your eyes. An agent needs your signature.
The bottleneck moves again
Simon’s logic did not retire in 1971. It repeats one level up. The internet made information cheap, and attention became the bottleneck. AI makes cognition cheap: drafting, comparing, summarizing, planning, all of it close to free. So the bottleneck moves again. The scarce resource is no longer time spent looking. It is judgment spent deciding.
Because AI does not just produce things for you to look at. It acts for you. It writes the email, books the flight, answers the customer, files the claim, moves the money. And software that acts for you needs something different from you. The old software wanted your time. Agents want your sign-off.
The workflow inverts with it. You used to think, act, and check. Now the machine thinks, proposes, and acts, and what remains for you is the approval in the middle and the accountability at the end. You stop operating and start governing.
Delegation keeps hitting walls a machine cannot cross alone. Is this my tone? Is the cheaper flight worth the 6 a.m. departure? Can it spend this much without asking? Who answers if this goes wrong? These are not information problems. A model can generate ten options in a second. It cannot want one of them.
Seconds, not hours
This is attention renting: a system borrowing a slice of your presence because it needs your intent, your taste, your permission, or your responsibility before it can continue. It does not want all of you, and it does not want you for long. It wants thirty seconds of your executive function, and then it wants you gone.
Capture wanted hours of low-value attention. Renting wants seconds of high-value judgment. The sessions get shorter. The stakes get higher. A feed could waste your evening. An agent can sign in your name.
The attention economy is not ending. It is compressing into the moments where consequences live.
The product that wins by leaving
This breaks the core assumption of internet product design. In the capture era, every extra minute you spent was revenue. In the renting era, every extra interruption is a tax. A product that keeps asking is a product that keeps failing.
So the metrics invert. Not time spent but time returned. Not engagement but correct escalation. Not daily active users but successful absences. The best agent handles what you would have handled and interrupts only when you would have wanted to be interrupted. Its highest achievement is not addiction. It is relief.
Call it trusted absence: the product earns the right to not be thought about. It is a strange thing to build a business on, which is why most of the industry will struggle to build it. The muscle memory runs the other way.
The rubber stamp
There is an obvious failure mode, and we have already run the experiment. Europe wrapped the web in consent banners and produced a continent that clicks accept all without reading. When oversight is designed as friction, people do not develop judgment. They develop a reflex.
Aviation found the same thing decades earlier. Autopilot did not remove pilots. It turned flying into supervising, and supervising quietly eroded the skills it was supposed to back up. Instructors called the new generation children of the magenta line: pilots who could follow the automation but could no longer confidently fly the plane.
Expect the same with agents. The twentieth approval of the day gets the same half second as the first. A human can be formally in the loop and practically asleep inside it. Oversight that consumes attention without producing understanding is not governance. It is liability theater. The signature is real. The judgment behind it is not.
Renting your weakest moment
Now the darker part. Capture manipulated what you saw. Renting can manipulate when you decide.
An agent that runs your life knows when you are rushed, tired, distracted. It controls the timing of the ask, the framing of the options, the default that wins if you do nothing. A system that wants a yes does not need to argue for it. It can wait for the moment you are least able to say no.
Social media optimized content for your cravings. Agent platforms can optimize the moment of consent itself. That is a more intimate kind of power, because the thing being engineered is no longer your taste. It is your agreement.
The harder question
For twenty years, consumer software was built on one question: how do we capture more attention? The answers gave us infinite scroll, autoplay, and the red badge, and they made trillions.
Agents replace it with a harder question: how do we deserve the attention we rent?
Capture only had to be irresistible. Renting has to be worth it. Every interruption now carries a claim: this moment needs your judgment more than whatever it pulled you from. Most products cannot make that claim honestly. The ones that can will be quiet, rare, and easy to forget you have.
Capture never answered for the hours it took. Renting should answer for every second.