AI agents are deciding who captures the value. Shoppers, workers and voters should have a say

Ask an AI assistant to buy you a coffee machine on Amazon and you have stepped into one of the first legal fights over agentic commerce. Perplexity’s Comet browser can browse the store and place the order for its user. Amazon sued, arguing that the agent entered its store without authorization while disguised as an ordinary Chrome browser, and a federal judge in San Francisco blocked it. On Aug. 4, the Ninth Circuit vacated that order. Because the agent acts only when its user tells it to, the court wrote, “it is the user who ‘accesses’ Amazon’s computers.” In September the court declined Amazon’s request for a rehearing. The underlying case continues. The dispute matters to shoppers because it is about what they get to see. Search Amazon yourself and the results page mixes sponsored listings with organic ones, flagged in small type, and you can scroll past them. Send an agent and you get back a short list with a sentence of explanation. You do not see the products it dropped, and you cannot tell whether its ranking reflects quality or what a seller paid for placement. Perplexity framed its defense in those terms, accusing Amazon of being “more interested in serving you ads, sponsored results, and influencing your purchasing decisions.” Amazon’s position is that a store may decide which software comes through its door. Both companies are fighting over the same asset, which is control of the moment a shopper makes up their mind. The price itself is becoming as hard to read as the ranking. In December 2025, Consumer Reports and Groundwork Collaborative asked 437 volunteers to buy the same groceries through Instacart. At one Safeway in Washington, D.C., a dozen eggs came up at $3.99, $4.28, $4.59 or $4.79 depending on the shopper. Instacart described the tests as limited and randomized, said they used no personal data, and ended them within two weeks of the report. The gap between the lowest and highest price for the same eggs was value moving from shoppers to the seller, and only the seller could see all of it. Cases like these are usually filed under consumer protection. They point to something larger. Each of us meets agentic AI as a consumer, as a worker and as a citizen, and in each role the same question is being settled: who captures the value that agents create. The public debate on AI and the economy has concentrated on productivity and jobs, on how much value AI will add. Far less attention has gone to how that value is divided, which depends on who controls the terms agents operate under. For workers, the division can already be measured. Researchers at the University of Oxford, working with the nonprofit Worker Info Exchange, analyzed more than 1.5 million trips by 258 Uber drivers in Britain between 2016 and 2024. After Uber introduced a dynamic pricing algorithm in 2023, passengers paid more per trip while drivers’ inflation-adjusted hourly earnings fell from over £22 to just over £19, before costs. Uber’s share of fares rose from around 25% to 29%, and on some trips exceeded half. Uber said it did not recognize the figures. The mechanism deserves attention whatever the exact numbers. One system sets what the passenger pays and what the driver earns, and it sees both sides of every trip. The passenger and the driver each see a single figure, and neither can tell how the other was priced. For citizens, the stakes are institutional. Friedrich Hayek’s defense of the market rested on the idea that prices emerge from countless decisions and belong to no one. The French philosopher Jean-Pierre Dupuy observed that this anonymity made market outcomes bearable, because a price set by no one offers no address for a grievance. Agentic pricing gives the price an address. When the Justice Department sued RealPage, it treated the vendor of the software that landlords used to set rents as a party to the coordination, alongside the landlords themselves. The proposed settlement announced in November 2025 bars RealPage from feeding its models competitors’ current nonpublic data and from training them on active leases. The rent on an apartment turned out to have an author. Inside companies, that author is harder to find, a problem two of us call “functional innocence” in our book IA et revenu (Ellipses, 2026): responsibility disappears across a chain of individually defensible decisions. Everyone involved can say in good faith that they decided nothing, which is why someone has to be named who answers for what the system does. These questions are being settled in court dockets and in the integration meetings where companies plug into assistants, and neither was designed for public deliberation. Where elected bodies have stepped in, the results have been modest and workable. New York now requires businesses that set prices with personal data to state, next to the price, “This price was set by an algorithm using your personal data.” In October 2025 a federal judge dismissed the National Retail Federation’s First Amendment challenge to the law. In Europe, the Platform Work Directive obliges platforms to tell workers when automated systems monitor them or make decisions about them, and member states must transpose it by Dec. 2, 2026. In July the European Commission ordered Google to open Android features to rival AI assistants. Disclosure is a start, and a real debate would go further using tools that already exist. The German economist Walter Eucken argued that competition is an order that has to be instituted and maintained, because a market left to itself drifts toward concentration. Applied to agents, that principle gives consumers a right to send the agent of their choice to any store they may lawfully enter, and to leave that agent taking with them what it has learned about them. The Ninth Circuit has addressed the first half of that right under a single statute, on a preliminary record. The principle also implies a duty of loyalty for agents that buy on our behalf. European law settled the equivalent question for financial advice in 2014: an adviser who claims to be independent may not keep commissions from the providers whose products it recommends. An agent that shops for you should meet the same standard, or say plainly who pays it. Business leaders have good reasons to want this debate. Most companies sit on the seller’s side of an agent’s shortlist, and they gain when channels stay open and no single assistant can set the terms of visibility alone. Executives are consumers, employees and voters as well. The question of who captures the value of agentic AI is being settled now, one ruling and one integration at a time, and leaving it there means accepting answers that few boards and no voters were asked about. The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.