First-hand information for everyone
This paper argues that digital search is moving away from the old model where people type keywords and scan ranked lists of links. In the new model, users state goals in ordinary language and software agents—small programs that act on a user’s behalf—interpret those goals, compare options, and return a recommendation or carry out the action. That shift turns search from a visible list you inspect into an embedded part of a system that can decide for you.
The authors describe this new architecture as an “agentic system.” They break it down into familiar parts: users; consumer agents that act for users; discovery layers that find and rank options; firms that provide goods and services; and service agents that act for firms. In practice, a user might say “find me a hotel like the one I stayed at last month but closer to downtown,” and a consumer agent would translate that request into actions instead of showing a ranked page of links.
Why this matters: the rules that govern competition, business models, and measurement change when agents mediate choices. In the old “attention economy,” firms competed for clicks and visibility. In the coming “preference economy,” competition will center on how well agents infer and satisfy user preferences. That means small design decisions—how agents access information, how options are surfaced, and how actions are executed—can have large effects on efficiency, competition, and consumer welfare. The paper notes early support from experimental agent-based marketplaces and economic theory for these claims.
The shift also changes how we measure success and build trust. Legacy metrics like relevance, click-through rates, dwell time, and conversion were built for a world where users inspect results directly. When agents act on behalf of users, measurement must help verify that the agent’s actions truly match the user’s preferences. The authors argue that auditability and observable alignment are not optional: without them, people may not trust agents enough to delegate decisions, and the efficiency gains of delegation will not be realized.
The authors stress caution. The field is still emerging, and many questions remain open. Market structures could evolve in different ways: embedded advertising, subscriptions, open or decentralized systems, or new hybrids are all possible. The paper calls for prospective research and careful system design at the intersection of artificial intelligence, economics, and system engineering. In short, the future of search will depend less on better ranking and more on how we design open, transparent, and competitive agentic systems—and on whether those systems can be audited and trusted.