The Intention Economy
The app economy competed for your attention. The next economy will compete for something more valuable: permission to act in your name.
Around 1905, in the operating room of a telephone exchange in Seattle, rows of young women sat before walls of blinking jacks and cords, performing a job that has now vanished so completely we forget it ever existed: they were the interface. You did not dial a number in 1905; you picked up the receiver and a human voice said “Number, please,” and that voice knew things. The operator knew who you were calling and how often. She knew the doctor’s line was busy but that he was actually next door at the pharmacy. In small towns she knew who was feuding with whom, and could, if she chose, connect you a little slower, or mention to the next caller that you had seemed agitated. The telephone company understood perfectly well that this human layer was both its greatest asset and its greatest liability, and so, according to one of the industry’s founding legends, did a Kansas City undertaker named Almon Strowger, who became convinced in the late 1880s that the local operator (allegedly the wife of a rival undertaker) was redirecting calls meant for his funeral parlor to her husband’s. Strowger’s revenge was the automatic telephone exchange, patented in 1891 and marketed, unforgettably, as the “girl-less, cuss-less telephone.” The dial tone that replaced the operator was more than a convenience. It was a promise of neutrality: no intelligence between you and the person you were trying to reach, no intermediary with knowledge of your affairs and interests of her own. It took the better part of a century to fully deliver that promise. It has taken about two years to begin revoking it, because the AI agent now being installed between every person and the digital world is, quite precisely, the return of the switchboard operator. Except this one remembers everything, serves a trillion-dollar company, and is about to handle not just your calls but your purchases, your plans, your information, and your choices.
From destinations to intermediaries
To see what is coming, imagine opening your phone five years from now and finding no grid of applications: no banking icon, no travel folder, no cluster of delivery services, no social platform demanding to be opened. There is simply an intelligent system that knows which services exist and can operate them when needed. You tell it you want to visit Paris in October; that you prefer an overnight flight; that Thursday’s meeting needs to move; that you would rather stay in a quiet neighborhood than near the monuments. The agent reads your calendar, compares flights, weighs hotels, checks the train from the airport, notices a restaurant you would probably love, and returns for your approval before booking anything. At no point did you consciously choose a company. You expressed an intention, and the digital world rearranged itself around it. This is the future implied by the movement from applications to agents, and it is far more than an upgrade in user experience; it is an inversion of the economy’s basic geometry. The app economy was organized around destinations: a company fought to place its icon on your first home screen because opening that icon meant entering its territory, where it controlled the shelves, framed the choices, and worked to convert your visit into a habit. The intention economy will be organized around intermediaries. The airline still flies the plane and the bank still holds the money, but if you never visit their interfaces again, they have been demoted from destinations to suppliers: interchangeable vendors standing in a warehouse behind a single counter, hoping the clerk calls their name.
The geometry inverts
App economy: you operate every service. Intention economy: one agent operates them for you.
Before the warnings begin, honesty requires me to show my hand: I want this future. I want it with an impatience that occasionally embarrasses the critic in me. I am tired of forms, of passwords, of hold music, of the seventeen small administrative interfaces standing between any intention and its completion, and on the day the agent genuinely works I will hand it my calendar with relief. The critique that follows is therefore not a rejection. It is the fine print I want read aloud before I sign, written by someone who fully intends to sign.
The most valuable position in this economy will therefore belong to whoever interprets human intention, and it is worth being unsentimental about how valuable that position is. If one system becomes the place where a few billion people begin their decisions, it will hold more influence over what gets discovered, considered, and bought than any institution in commercial history: more than the great department stores held over merchandise, more than the television networks held over evening opinion, more than the search engine holds today over the visible internet. We have watched this movie in miniature. Search engines organized the web by ranking pages, and entire industries learned to live and die by their position on one results screen. Social platforms organized culture by ranking posts, and publishers, musicians, and politicians reshaped their output to please the feed. AI agents would organize something categorically larger, not what you see but what is done: which airline is booked, which insurer is quoted, which article is summarized into your morning briefing and which quietly is not. And if the history of digital gateways teaches anything, it is that they do not stay plural. Each layer of the modern stack that began as an open field has resolved into one or two chokepoints commanding nearly the whole territory.
Gateways concentrate
Share of each digital gateway held by its top players (global, approximate)
The stakes of the stack
This is the light in which to read the noisy debate between open and closed AI models, which is usually framed as a technical disagreement about performance, safety, and software licensing but is at bottom a disagreement about who will own the infrastructure through which society thinks and acts. Closed systems offer real virtues (coherence, reliability, someone identifiable to blame), but they ask users to trust the values, commercial incentives, and invisible rankings of their owners, indefinitely, on faith. Open systems permit inspection, experimentation, and local control, but openness alone does not guarantee accountability; a model with published weights can still be deployed behind an interface that discloses nothing. The genuinely political question, the one that will outlast every benchmark, is whether individuals and institutions retain meaningful alternatives (whether switching remains possible, whether the exits stay unlocked) or whether convenience gradually consolidates authority inside a handful of planetary systems, as it did in every previous layer of the stack. For what it is worth, my own position, revised more than once in the writing of this essay and held loosely still, is that the real fault line is not open versus closed but exits versus no exits: I will accept a closed, polished agent on the day I can leave it without losing my history, and not one day earlier. We have seen the general pattern before, and not only in software. Nineteenth-century factories centralized production because industrial machinery demanded capital no craftsman could raise. Newspapers and broadcasters centralized attention because presses and transmitters were expensive. Digital platforms centralized social life through network effects that made leaving costly precisely in proportion to how many others had joined. Each system began by offering access. Each ended by setting terms. AI agents could extend the series one more step, centralizing not production, nor information, nor social connection, but the raw upstream resource from which all of those flow: intention itself. And the stakes climb further as intelligence leaves the screen, because the same models are now being fitted into vehicles, robots, and buildings: a world in which AI is no longer a destination you visit but the environment through which you move, where the store interprets you as you interpret its shelves, and the already-blurry boundary between online and offline finishes dissolving.
It must be said, and not grudgingly, that this future could be remarkably humane, and the case for it deserves its full weight. A well-designed agent could dissolve the bureaucratic cruft that consumes whole days of modern existence: the forms, the hold music, the seventeen-step reimbursements. It could hand a two-person business the operational reach of a corporation, help a patient actually understand a diagnosis, walk a first-generation student through an application process built by and for insiders, and let an eighty-year-old operate systems that currently punish anyone who did not grow up inside them. The deep promise of AI is not that it automates labor; it is that it could abolish the penalty civilization quietly imposes on people for not already understanding its systems, a penalty that has always fallen hardest on the poor, the old, the sick, and the new. The switchboard operator, remember, was also a service: she found the doctor at the pharmacy. The question was never whether an intermediary can help. It is what the intermediary knows, whom she works for, and what she does with the knowing.
Which is why the arriving era needs a distinction we have never before had to draw sharply: the difference between delegation and surrender. Delegation is revocable, inspectable, and bounded. You know when the agent is making a consequential choice, which interests shaped it, and how to challenge it; you can see what the system remembers about you, govern what it shares, and carry your history to a competitor the way you carry a phone number today. Surrender is what happens when none of that is true and nobody noticed, when consent means having clicked “accept” once, years ago, on a screen designed to be dismissed, while an invisible system accumulated authority one convenient transaction at a time. The app economy taught us that convenience is never free: we paid in attention, in data, and in dependency, and mostly we paid without reading the price. The intention economy will ask for something categorically greater: standing permission to act in our name. That permission can produce enormous good, but it is not a feature to be buried in settings; it is a power of attorney, and it deserves to be treated with the gravity that phrase implies. The defining question of the next decade will not be whether artificial intelligence can carry out our intentions; it plainly, increasingly, can. The question is whether, after we have delegated enough of them, we will still understand the decisions being made on our behalf, still know how to ask what the operator knows, and still remember, as Almon Strowger did, that when the intermediary’s interests and ours diverge, the answer is not to stop calling. It is to insist on a system where we can see the switchboard.
Notes & sources: the app-to-agent argument follows discussions at the RAISE Summit (Paris); Almon Strowger’s automatic exchange, patented 1891 (the rival-undertaker story is industry legend, printed since the 1900s); photo: Sunset Telephone Company operators, Seattle ca. 1905, public domain via Wikimedia Commons/MOHAI; gateway shares are approximate global figures from StatCounter and industry ad-revenue estimates; historical centralization framing after standard histories of industrialization and broadcasting.