The Human in the Loop Is Leaving the Building
Ari Emanuel and Ben Thompson on why AI agents will break Google and Amazon's ad empires—and why live events are what's left.
By The ai-compute Desk
Ari Emanuel called Ben Thompson out of the blue—the way he always does—and within thirty seconds the conversation had turned to the death of search advertising, the future of social networks, and why the generation of human desire is the only business that AI cannot touch. [1] The result is less interview than sparring match, two people who have spent years thinking about media and money arriving at the same uncomfortable conclusion from opposite directions.
Emanuel is Executive Chairman of WME Group, Executive Chair and CEO of TKO Group Holdings—which owns UFC and WWE—and founder of MARI, a live events and experiences company. [1] TKO's portfolio reaches more than one billion households across 210 countries and territories and organizes more than 500 live events per year, attracting more than three million fans. [2] That scale is not incidental to the argument Emanuel and Thompson are making; it is the argument.
The conversation turns on a single pivot: what happens to money when the intermediary disappears. Every industry that built its economics around capturing attention in transit—between a search query and a purchase, between a cable bundle and a sports broadcast—now finds that transit route collapsing. What endures is the moment itself: the crowd, the fight, the spectacle that cannot be compressed into a prompt.
Thompson's core claim is architectural. When ChatGPT arrived, he argues, it was still a threat Google could absorb—because the human remained in the loop. [1] A person typed a query, skimmed results, clicked an ad. But AI agents change the geometry entirely. [1]
Amazon is next. [1] Amazon's physical logistics network—warehouses, trucks, last-mile delivery—has always been described as a moat. Thompson reframes it as a liability: "The problem with having a physical moat is that means you have a lot of capital tied up in that, which means you do have a forcing function that means you have to adjust." [1] A digital competitor can pivot overnight; Amazon cannot.
The advertising model that replaces the search-skim model is, in Thompson's telling, the inverse of surveillance capitalism. Instead of matching user profiles against product databases—what he calls the two-towers approach—the next generation of ad targeting will run a personalized prediction engine inside a user's own virtual machine. [1] The system predicts what a person wants before they know they want it, and it does so privately, without the data ever leaving the device. "They're going to be credibly able to say it's totally private, because it is totally private," Thompson said. [1] The irony is that this model is more powerful than anything that came before, and it does not require selling user data at all.
Thompson connected this to Meta's Muse. When Muse launched, Meta promised it would not use Muse data for advertising—a privacy commitment that Thompson argues was a strategic miscalculation. "No one knows they made that promise, and no one cares, because no one cares about that stuff," he said. [1] The promise constrained Meta's options without buying goodwill. Emanuel drew an immediate parallel to Apple: "Just like Apple, we're not going to look at your data, major mistake when Tim [Cook] said that." [1] Both companies optimized for a privacy narrative at the precise moment when privacy was about to become technically enforceable rather than merely promised.
There is something almost counter-intuitive about where all of this lands. The more sophisticated the prediction engine, the more precisely it can model desire—and yet the surfaces that capture the value of that desire are increasingly analog. Not a feed, not a results page, but a room full of people watching something that cannot be rewound.
This is Emanuel's bet with MARI, and with TKO more broadly. The thesis is simple: live events cannot be disintermediated. An AI agent can find you a cheaper flight, cancel your lingering subscriptions, and route around every search ad between you and a purchase. [1] It cannot replicate the experience of being present. Emanuel cited Elon Musk making exactly this point to him directly—that live, physical experience sits outside the reach of algorithmic substitution. [1] Whether or not that attribution carries weight, the structural logic holds.
The cable era is the cautionary precedent. [1] "And once that broke with Netflix, etc., it just went away. The cable operators and cable were broken, it just fell to the ground," he said. [1] Sports fans, Thompson observed, were the biggest welfare recipients in history throughout the entire cable era. Now they pay directly, but advertisers will pay premiums for that concentrated, high-intent audience on whatever platform carries the rights.
Social networks occupy a complicated middle position in this framework. [1] Emanuel raised Snapchat separately as one of the last social networks, still possessed of a great user base. Thompson noted that the engagement is off the charts—and that it remains a clearly tremendously valuable asset, though Snapchat remains an open question. [1]
TikTok's ascent reframes the entire social graph debate. Thompson's analysis is that TikTok is not, technically, a social network at all—it is a user-generated content network whose algorithm finds the best content across the entire service and surfaces it regardless of social connection. [1] "The vast majority of content sucks, but the absolute value of good content is very large, because the total amount of content is so great, even if you get a very small number on a massive denominator," he said.
[1] But the conversation treats phone calls as more than personal habit—Thompson argues that voice communication is high-bandwidth in a way text cannot match. You get the shape of a problem and the shape of a person simultaneously. In an era when AI can generate plausible text responses indefinitely, the irreducibly human signal—tone, hesitation, the half-finished thought—may become more valuable, not less. Emanuel, who has been dyslexic his whole life and has always preferred the phone, arrived at this conclusion by a different route than Thompson, but they end up in the same place.
What the conversation ultimately sketches is a map of value after the great deflation. Computation gets cheaper. Prediction gets sharper. Distribution collapses into a handful of surfaces controlled by a handful of companies. And the things that remain scarce—the crowd, the call, the room where something actually happened—acquire a gravity that the abundance economy cannot simulate away.
[1] Emanuel thinks it goes higher still. The conservative scenario, Thompson argues, is that Google and Amazon advertising are both impaired by agent-mediated commerce, that money flows to surfaces not touched by AI, and that live events and social networks with genuine human participation absorb the bulk of what shifts. [1] The aggressive scenario is that the entire advertising economy reconfigures around prediction, presence, and the irreducible fact of being in the room.
Emanuel built his career on rolling calls—on the premise that personal contact compounds in ways that impersonal contact does not. The irony is that the AI economy he and Thompson are describing has arrived at the same conclusion from the opposite direction: not because phone calls are nostalgic, but because they are one of the last things a language model cannot convincingly fake. [1]