The Agentic Economy DOI: XXXXXXX.XXXXXXX Journal: JACM Volume: 37 4 111 8 CCS: Social and professional topics Economic impact
David M. Rothschild, Markus Mobius, Jake M. Hofman, Eleanor Dillon, Daniel G. Goldstein, Nicole Immorlica, Sonia Jaffe, Brendan Lucier, Aleksandrs Slivkins, Matthew Vogel · 2025
Read The Agentic Economy DOI: XXXXXXX.XXXXXXX Journal: JACM Volume: 37 4 111 8 CCS: Social and professional topics Economic impact by David M. Rothschild, Markus Mobius, Jake M. Hofman, Eleanor Dillon, Daniel G. Goldstein, Nicole Immorlica, Sonia Jaffe, Brendan Lucier, Aleksandrs Slivkins, Matthew Vogel (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-15799.txt.
Introduction
Personal AI agents that shop, book, and bargain for their owners could reshape markets far more than the solo productivity gains seen so far. David Rothschild and co-authors map that shift in a 2025 paper on the agentic economy. In their picture, buyer agents and seller agents meet directly, talking in plain language through shared protocols.
The authors split two questions: whether agents can talk freely, a technical matter, and whether markets will let them, a matter of rules and power. Today's agents sit trapped inside single platforms or run one firm's whole chain. Futures range from walled gardens to an open web of agents. Ads, discovery, tiny payments, and the bundling of digital goods all stand to change. The design of agent talk, they argue, will decide who gains.
Catalog record
| CREATOR | David M. Rothschild, Markus Mobius, Jake M. Hofman, Eleanor Dillon, Daniel G. Goldstein, Nicole Immorlica, Sonia Jaffe, Brendan Lucier, Aleksandrs Slivkins, Matthew Vogel |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 20 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-15799
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
The Agentic Economy DOI: XXXXXXX.XXXXXXX Journal: JACM Volume: 37 4 111 8 CCS: Social and professional topics Economic impact David M. Rothschild, Markus Mobius, Jake M. Hofman, Eleanor Dillon, Daniel G. Goldstein, Nicole Immorlica, Sonia Jaffe, Brendan Lucier, Aleksandrs Slivkins, Matthew Vogel Abstract . 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 and service agents interact 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. 1. Introduction Generative AI has revolutionized the way we interact with technology, allowing people to express their intent in free-form natural language. It has paved the way for AI agents that not only converse with users but also perform actions on their behalf, flexibly and with minimal guidance. Delegation to AI has already begun to improve the efficiency of individual processes , making both consumers and businesses more productive in the set of tasks they had already been doing ( brynjolfsson2025generative ; cui2024effects , e.g.,) . However, we believe that the more disruptive—and yet to be realized—impact of generative AI is its potential to drastically reduce the communication frictions between (and among) consumers and businesses . This could lead to a reorganization of markets, shifts of market power, and the introduction of entirely new products and services. Consumers have traditionally faced high communication costs when initiating relationships with businesses, reducing efficiency ( klemperer1995competition ) . For example, a consumer seeking a new tax preparer might hesitate to switch because she would have to explain her financial situation all over again to a new person or online service. These communication hurdles can prevent consumers from taking advantage of better products and services or lower prices. Businesses have tried to lower these costs with tools like online forms and voicemail menus, but these often just shift communication costs to the consumer and can make interactions more rigid. Imagine instead a future where every consumer has an assistant agent to communicate their preferences and personal information to businesses, and every business has service agents to interact with consumers and other businesses. These agents could be designed to interface with each other seamlessly and flexibly, transforming the landscape of consumer-business interactions. Delegating interactions to such assistant and service agents lowers communication costs, expanding the landscape of options available to both consumers and businesses, increasing the efficiency of markets. To unlock the full economic potential of generative AI’s communication capabilities, two developments are necessary. First, consumers and businesses must widely adopt assistant and service agents. This is already underway ( bick2024rapid ; humlum2024adoption ) . Second, these agents must be designed to programmatically interact with each other to facilitate transactions. On the technical front, there has been significant progress in standardizing such agentic interaction, with frameworks like Microsoft’s AutoGen ( MicrosoftAutogen ) , and protocols like Anthropic’s Model Context Protocol ( AnthropicMCP ) and Google’s Agent2Agent Protocol ( GoogleA2A ) . However, it remains to be seen how these advances will be adopted and implemented, or constrained, given their complex interplay with and dependence on market forces. The largest benefits of inter-agent communication will be realized as markets reorganize around these new capabilities ( BRESNAHAN1995 ) . While the possibilities of this technological distribution are numerous, we offer a framework for understanding the most plausible outcomes and contrasting them with the current state. The key question is whether inter-agent communication will occur within closed "agentic walled gardens" controlled by a few dominant providers, akin to today’s app stores, or through a more open "web of agents" where agents freely connect and transact, much like today’s World Wide Web. The answer to this question will determine how today’s largest online platforms are impacted by the proliferation of interconnected agents, whether it be further entrenchment or diminished market dominance and democratization of the economic benefits of AI technology. We explore how those benefits might manifest, with implications for advertising, e-commerce, and the creation of new products and industries. 2. The Current State of Agentic AI Before discussing future possibilities for agentic economies, it is helpful to survey the current landscape of AI agents. On the surface it may appear as if several existing efforts are well on their way to providing consumers and businesses with assistant and service agents that could function as described above. But most existing agents lack a key ingredient: while they are designed to interact with or simulate human users, few public offerings are designed to interact with each other. Existing agents generally come in one of two forms: siloed service agents or general-purpose end-to-end agents. The first, siloed service agents, provide a new user interface for products and services within a single company. For example, Amazon’s Rufus allows customers visiting Amazon to access their order histories or compare potential purchases by features through natural language instead of navigating a website. Likewise, Expedia’s Romie provides a chat interface to help customers build travel itineraries—including flights, hotels, and restaurants—by pulling information from customer emails and group chats. Importantly, however, these efforts do not expose interfaces intended for interaction with other agents. As a result, they are still fundamentally limited by people having to navigate to and personally interact with them. End-to-end agents take a different approach, aiming to provide general-purpose automated functionality that is not limited to a single company or service. For instance, technology from OpenAI, Google, and Microsoft can aggregate research from external sources, navigate business websites on the user’s behalf, and even perform simple tasks like making reservations or placing food orders, all within one interface. Importantly, however, much of the functionality that these end-to-end agents provide currently comes through “computer use models” that simulate a user pointing and clicking on existing (non-agentic) websites. This gives the illusion of assistant and service agents working together, but the absence of a true service agent on the business side limits what an end-to-end agent can accomplish. Moreover, by mimicking human users, these agents risk creating adversarial relationships with businesses that, for example, rely heavily on advertising revenue and may resist having their websites accessed by agents instead of humans. Even if end-to-end agents can perfectly simulate human users and businesses agree to agent-based access, interactions would still be constrained by what businesses currently expose