Digital search is undergoing a fundamental transformation from a human-driven process of discovery to an agent-mediated system of delegated decision-making. In the traditional model of digital search, users translate intent into keyword-based queries, evaluate ranked lists of links, and execute decisions outside the search interface. In an AI-native world, users express goals in natural language, agents interpret these intentions, and outcomes are returned as recommendations or executed decisions. This shift moves search from a link-based user interface to an embedded system component, with implications for transparency, competition, and monetization. The resulting system design problem raises key questions about information quality and access, trust, incentive alignment, and market structure. Early evidence from experimental agent-mediated marketplaces and economic theory suggests that small design choices, such as how stakeholders access information, how options are surfaced, and how actions are executed, have first-order effects on efficiency, competition, and the welfare of consumers and firms. We propose that the future of search will be determined not by incremental improvements in ranking algorithms and natural-language interfaces, but by the design of open, transparent, and competitive agentic systems that govern how decisions are made and how markets operate, highlighting a set of grand challenges at the intersection of AI, economics, and system design.

Motivated by agentic markets – two-sided markets in which consumers and businesses are assisted by AI tools that facilitate consumers' search – we study the impact of improved search technology on learning and welfare in markets. We put forth a model where consumers engage in costly search to acquire signals of product fit prior to purchase. The market tracks indications of fit for searched products and indications of quality for chosen products, thereby guiding searches. We characterize the long-run steady-state of the resulting dynamics as well as the impact of improving search technology. We find cheaper search improves learning and consumer surplus, whereas more informative search can degrade both unless the market learns as much as consumers about the products by, for example, “reading the transcripts” of agentic conversations. Finally, we consider the impact of search improvements on how businesses set prices. At equilibrium prices in symmetric markets, consumer surplus is improved by cheaper search but may be decreased by more informative search, due to weakened inter-business competition.

Artificial intelligence feels omnipresent, yet the disruption many expect has not fully arrived. The main reason is not model capability, nor even the tools built to harness those models. Rather, most organizations are still using AI to accelerate workflows designed for a pre-AI world. We offer a three-stage lens: Augmentation, Automation, and Reconstruction, and argue that the most consequential disruption resides in the third stage where workflows and markets are rebuilt around delegation, machine-to-machine interaction, continuous monitoring, and auditable constraints. Achieving this system-level transformation takes time: it requires trust and accountability infrastructure, machine-legible and interoperable data and interfaces, the design and adoption of these new workflows, and economic incentives that favor reconstruction rather than local optimization: the complementary investments that produce the familiar "productivity J-curve" of general-purpose technologies. We illustrate this transition through examples in consumer markets, education, news, and coding. Finally, we emphasize a normative point: the agentic future is not predetermined. Leaders must both skate to where the puck is going and actively steer it toward a good place, ensuring innovation delivers welfare gains felt by businesses and consumers around the world.

Generative AI has transformed human-computer interaction by enabling natural language interfaces and the emergence of autonomous agents capable of acting on users' behalf. While early applications have improved individual productivity, these gains have largely been confined to predefined tasks within existing workflows. We argue that the more profound economic impact lies in reducing communication frictions between consumers and businesses. This shift could reorganize markets, redistribute power, and catalyze the creation of new products and services. We explore the implications of an agentic economy, where assistant agents act on behalf of consumers and service agents represent businesses, interacting programmatically to facilitate transactions. A key distinction we draw is between unscripted interactions – enabled by technical advances in natural language and protocol design – and unrestricted interactions, which depend on market structures and governance. We examine the current limitations of siloed and end-to-end agents, and explore future scenarios shaped by technical standards and market dynamics. These include the potential tension between agentic walled gardens and an open web of agents, implications for advertising and discovery, the evolution of micro-transactions, and the unbundling and rebundling of digital goods. Ultimately, we argue that the architecture of agentic communication will determine the extent to which generative AI democratizes access to economic opportunity.

As LLM agents advance, they are increasingly mediating economic decisions, ranging from product discovery to transactions, on behalf of users. Such applications promise benefits but also raise many questions about agent accountability and value for users. Addressing these questions requires understanding how agents behave in realistic market conditions. However, previous research has largely evaluated agents in constrained settings, such as single-task marketplaces (e.g., negotiation) or structured two-agent interactions. Real-world markets are fundamentally different: they require agents to handle diverse economic activities and coordinate within large, dynamic ecosystems where multiple agents with opaque behaviors may engage in open-ended dialogues. To bridge this gap, we investigate two-sided agentic marketplaces where Assistant agents represent consumers and Service agents represent competing businesses. To study these interactions safely, we develop Magentic-Marketplace– a simulated environment where Assistants and Services can operate. This environment enables us to study key market dynamics: the utility agents achieve, behavioral biases, vulnerability to manipulation, and how search mechanisms shape market outcomes. Our experiments show that frontier models can approach optimal welfare– but only under ideal search conditions. Performance degrades sharply with scale, and all models exhibit severe first-proposal bias, creating 10-30x advantages for response speed over quality. These findings reveal how behaviors emerge across market conditions, informing the design of fair and efficient agentic marketplaces.

We study a communication game between a sender and receiver where the sender has access to a set of informative signals about a state of the world. The sender chooses one of her signals, called an ``anecdote'' and communicates it to the receiver. The receiver takes an action, yielding a utility for both players. Sender and receiver both care about the state of the world but are also influenced by a personal preference so that their ideal actions differ. We characterize perfect Bayesian equilibria when the sender cannot commit to a particular communication scheme. In this setting the sender faces ``persuasion temptation'': she is tempted to select a more biased anecdote to influence the receiver's action. Anecdotes are still informative to the receiver but persuasion comes at the cost of precision. This gives rise to ``informational homophily'' where the receiver prefers to listen to like-minded senders because they provide higher-precision signals. In particular, we show that a sender with access to many anecdotes will essentially send the minimum or maximum anecdote even though with high probability she has access to an anecdote close to the state of the world that would almost perfectly reveal it to the receiver. In contrast to the classic Crawford-Sobel model, full revelation is a knife-edge equilibrium and even small differences in personal preferences will induce highly polarized communication and a loss in utility for any equilibrium. We show that for fat-tailed anecdote distributions the receiver might even prefer to talk to poorly informed senders with aligned preferences rather than a knowledgeable expert whose preferences may differ from her own. We also show that under commitment differences in personal preferences no longer affect communication and the sender will generally report the most representative anecdote closest to the posterior mean for common distributions.

We introduce a model of online algorithms subject to strict constraints on data retention. An online learning algorithm encounters a stream of data points, one per round, generated by some stationary process. Crucially, each data point can request that it be removed from memory m rounds after it arrives. To model the impact of removal, we do not allow the algorithm to store any information or calculations between rounds other than a subset of the data points (subject to the retention constraints). At the conclusion of the stream, the algorithm answers a statistical query about the full dataset. We ask: what level of performance can be guaranteed as a function of m? We illustrate this framework for multidimensional mean estimation and linear regression problems. We show it is possible to obtain an exponential improvement over a baseline algorithm that retains all data as long as possible. Specifically, we show that m = Poly (d, log(1/epsilon)) retention suffices to achieve mean squared error. after observing O (1/epsilon) d-dimensional data points. This matches the error bound of the optimal, yet infeasible, algorithm that retains all data forever. We also show a nearly matching lower bound on the retention required to guarantee error epsilon. One implication of our results is that data retention laws are insufficient to guarantee the right to be forgotten even in a non-adversarial world in which firms merely strive to (approximately) optimize the performance of their algorithms. Our approach makes use of recent developments in the multidimensional random subset sum problem to simulate the progression of stochastic gradient descent under a model of adversarial noise, which may be of independent interest.

Do celebrity endorsements matter? And if so, how can celebrities communicate effectively? We conduct a nationwide Twitter experiment in Indonesia promoting vaccination. Celebrity messages are 72% more likely to be passed on or liked than similar messages without a celebrity’s imprimatur. In total, 66% of the celebrity effect comes from authorship, compared to passing on messages. Citing external medical sources decreases retweets by 27%. Phone surveys show that those randomly exposed to messaging have fewer incorrect beliefs and report more vaccination among friends and neighbours. The results can inform public health campaigns and celebrity public service more generally.

We use a series of experiments to understand whether and how people’s beliefs about their own abilities are biased relative to the Bayesian benchmark and how these beliefs then affect behavior. We find that subjects systematically and substantially overweight positive feedback relative to negative (asymmetry) and also update too little overall (conservatism). These biases are substantially less pronounced in an ego-free control experiment. Updating does retain enough of the structure of Bayes’ rule to let us model it coherently in an optimizing framework, in which, interestingly, asymmetry and conservatism emerge as complementary biases. We also find that exogenous changes in beliefs affect subjects’ decisions to enter into a competition and do so similarly for more and less biased subjects, suggesting that people cannot “undo” their biases when the time comes to decide. This paper was accepted by Axel Ockenfels, behavioral economics and decision analysis. Funding: Financial support from the National Science Foundation (NSF), Harvard University, and Wesleyan University is gratefully acknowledged. P. Niehaus received financial support from an NSF Graduate Research Fellowship. Supplemental Material: The data files are available at https://doi.org/10.1287/mnsc.2021.4294 .

Partisan segregation within the news audience buffers many Americans from countervailing political views, posing a risk to democracy. Empirical studies of the online media ecosystem suggest that only a small minority of Americans, driven by a mix of demand and algorithms, are siloed according to their political ideology. However, such research omits the comparatively larger television audience and often ignores temporal dynamics underlying news consumption. By analyzing billions of browsing and viewing events between 2016 and 2019, with a novel framework for measuring partisan audiences, we first estimate that 17% of Americans are partisan-segregated through television versus roughly 4% online. Second, television news consumers are several times more likely to maintain their partisan news diets month-over-month. Third, TV viewers’ news diets are far more concentrated on preferred sources. Last, partisan news channels’ audiences are growing even as the TV news audience is shrinking. Our results suggest that television is the top driver of partisan audience segregation among Americans.

Since the 2016 US presidential election, the deliberate spread of misinformation online, and on social media in particular, has generated extraordinary concern, in large part because of its potential effects on public opinion, political polarization, and ultimately democratic decision making. Recently, however, a handful of papers have argued that both the prevalence and consumption of "fake news" per se is extremely low compared with other types of news and news-relevant content. Although neither prevalence nor consumption is a direct measure of influence, this work suggests that proper understanding of misinformation and its effects requires a much broader view of the problem, encompassing biased and misleading-but not necessarily factually incorrect-information that is routinely produced or amplified by mainstream news organizations. In this paper, we propose an ambitious collective research agenda to measure the origins, nature, and prevalence of misinformation, broadly construed, as well as its impact on democracy. We also sketch out some illustrative examples of completed, ongoing, or planned research projects that contribute to this agenda.

The DeGroot model has emerged as a credible alternative to the standard Bayesian model for studying learning on networks, offering a natural way to model naïve learning in a complex setting. One unattractive aspect of this model is the assumption that the process starts with every node in the network having a signal. We study a natural extension of the DeGroot model that can deal with sparse initial signals. We show that an agent’s social influence in this generalized DeGroot model is essentially proportional to the degree-weighted share of uninformed nodes who will hear about an event for the first time via this agent. This characterization result then allows us to relate network geometry to information aggregation. We show information aggregation preserves “wisdom” in the sense that initial signals are weighed approximately equally in a model of network formation that captures the sparsity, clustering, and small-world properties of real-world networks. We also identify an example of a network structure where essentially only the signal of a single agent is aggregated, which helps us pinpoint a condition on the network structure necessary for almost full aggregation. Simulating the modeled learning process on a set of real-world networks, we find that there is on average 22.4 percent information loss in these networks. We also explore how correlation in the location of seeds can exacerbate aggregation failure. Simulations with real-world network data show that with clustered seeding, information loss climbs to 34.4 percent. (JEL D83, D85, Z13)

A policy debate centers around the question how news aggregators such as Google News affect traffic to online news sites.Many publishers view aggregators as substitutes for traditional news consumption while aggregators view themselves as complements because they make news discovery easier.We use Spain as a natural experiment because Google News shut down altogether in response to a copyright reform enacted in December 2014.We compare the news consumption of a large number of Google News users with a synthetic control group of similar non-Google News users.We find that the shutdown of Google News reduces overall news consumption by about 20% for treatment users, and reduces page views on publishers other than Google News by 10%.This decrease is concentrated around small publishers.We further find that users are able to replace some but not all of the types of news they previously read.Postshutdown, they read less breaking news, hard news, and news that is not well covered on their favorite news publishers.These news categories explain most of the overall reduction in news consumption, and shed light on the mechanisms through which aggregators interact with traditional publishers.

We study a model of social learning and communication using hard anecdotal evidence. There are two Bayesian agents (a sender and a receiver) who wish to communicate. The receiver must take an action whose payoff depends on their personal preferences and an unknown state of the world. The sender has access to a collection of n samples correlated with the state of the world, which we think of as specific anecdotes or pieces of evidence, and can send exactly one of these samples to the receiver in order to influence her choice of action. Importantly, the sender's personal preferences may differ from the receiver's, which affects the seller's strategic choice of which anecdote to send. We show that if the sender's communication scheme is observable to the receiver (that is, the choice of which anecdote to send given the set they receive), then they will choose an unbiased and maximally informative communication scheme, no matter the difference in preferences. Without observability, however, even a small difference in preferences can lead to a significant bias in the choice of anecdote, which the receiver must then account for. This can significantly reduce the informativeness of the signal, leading to substantial utility loss for both sides. One implication is informational homophily: a receiver can rationally prefer to obtain information from a poorly-informed sender with aligned preferences, rather than a knowledgeable expert whose preferences may differ from her own.Institutional subscribers to the NBER working paper series, and residents of developing countries may download this paper without additional charge at www.nber.org.

We examine potential bias in Facebook’s 10-trillion cell URLs dataset, consisting of URLs shared on its platform and their engagement metrics. Despite the unprecedented size of the dataset, it was altered to protect user privacy in two ways: 1) by adding differentially private noise to engagement counts, and 2) by censoring the data with a 100-public-share threshold for a URL’s inclusion. To understand how these alterations affect conclusions drawn from the data, we estimate the prevalence of fake news in the massive, censored URLs dataset and compare it to an estimate from a smaller, representative dataset. We show that censoring can substantially alter conclusions that are drawn from the Facebook dataset. Because of this 100-public-share threshold, descriptive statistics from the Facebook URLs dataset overestimate the share of fake news and news overall by as much as 4X. We conclude with more general implications for censoring data.

Surveys are a vital tool for understanding public opinion and knowledge, but they can also yield biased estimates of behavior. Here we explore a popular and important behavior that is frequently measured in public opinion surveys: news consumption. Previous studies have shown that television news consumption is consistently overreported in surveys relative to passively collected behavioral data. We validate these earlier findings, showing that they continue to hold despite large shifts in news consumption habits over time, while also adding some new nuance regarding question wording. We extend these findings to survey reports of online and social media news consumption, with respect to both levels and trends. Third, we demonstrate the usefulness of passively collected data for measuring a quantity such as "consuming news" for which different researchers might reasonably choose different definitions. Finally, recognizing that passively collected data suffers from its own limitations, we outline a framework for using a mix of passively collected behavioral and survey-generated attitudinal data to accurately estimate consumption of news and related effects on public opinion and knowledge, conditional on media consumption.

. Do celebrity endorsements matter? And if so, how can celebrities communicate effectively? We conduct a nationwide Twitter experiment in Indonesia promoting vaccination. Celebrity messages are 72 percent more likely to be passed or liked than similar messages without a celebrity’s imprimatur. Decomposing this, 79 percent of the celebrity effect comes from authorship, compared to passing on messages. Citing external medical sources decreases retweets by 27 percent. Phone surveys show that those randomly exposed to messaging have fewer incorrect beliefs and report more vaccination among friends and neighbors. The results can inform public health campaigns, such as for COVID-19 vaccination.

Although it is under-studied relative to other social media platforms, YouTube is arguably the largest and most engaging online media consumption platform in the world. Recently, YouTube's scale has fueled concerns that YouTube users are being radicalized via a combination of biased recommendations and ostensibly apolitical "anti-woke" channels, both of which have been claimed to direct attention to radical political content. Here we test this hypothesis using a representative panel of more than 300,000 Americans and their individual-level browsing behavior, on and off YouTube, from January 2016 through December 2019. Using a labeled set of political news channels, we find that news consumption on YouTube is dominated by mainstream and largely centrist sources. Consumers of far-right content, while more engaged than average, represent a small and stable percentage of news consumers. However, consumption of "anti-woke" content, defined in terms of its opposition to progressive intellectual and political agendas, grew steadily in popularity and is correlated with consumption of far-right content off-platform. We find no evidence that engagement with far-right content is caused by YouTube recommendations systematically, nor do we find clear evidence that anti-woke channels serve as a gateway to the far right. Rather, consumption of political content on YouTube appears to reflect individual preferences that extend across the web as a whole.