The agentic economy: How generative AI may disrupt digital intermediaries

Today’s platforms, Amazon, Spotify, OpenTable, and Expedia, serve as centralized intermediaries between consumers and businesses, standardizing discovery, search, and transactions. These two-sided platforms derive value from controlling the interfaces on both sides, often capturing substantial margins through referral fees and limited customization.

Generative AI is no longer just a productivity tool - it is becoming a market architect. A landmark study proposes that the most disruptive consequence of generative AI may not be task automation, but the reorganization of entire markets around autonomous agents.

Published on arXiv under the title “The Agentic Economy”, the study argues that if consumers and businesses adopt assistant and service agents capable of interacting directly with each other, traditional intermediaries may lose relevance, new ecosystems may emerge, and competition could be redefined across sectors.

Will AI Agents Displace Digital Intermediaries?

Today’s platforms, Amazon, Spotify, OpenTable, and Expedia, serve as centralized intermediaries between consumers and businesses, standardizing discovery, search, and transactions. These two-sided platforms derive value from controlling the interfaces on both sides, often capturing substantial margins through referral fees and limited customization.

The study warns that widespread deployment of autonomous agents on both consumer and business ends could reduce reliance on such platforms. When consumers delegate their preferences to assistant agents, and businesses deploy service agents capable of interpreting and responding to those agents, the need for a centralized marketplace may diminish. This would enable dynamic, unscripted interactions between buyers and sellers, disrupting the traditional logic of constrained interfaces and high switching costs.

While intermediary platforms will not disappear overnight, they still offer value in dispute resolution, reputation signaling, and compliance, the study anticipates fierce competition among platforms as agent-based discovery reduces the cost of switching. In turn, this may drive down platform margins and force a rethink of their economic role in digital commerce.

Will AI Economies Evolve as Open Webs or Walled Gardens?

The next major shift depends on how agentic interactions are governed. The authors outline two divergent futures: an “agentic web,” modeled on the open architecture of today’s Internet, and “agentic walled gardens,” similar to app ecosystems in mobile operating systems.

In an open agentic web, assistant and service agents would interact freely across organizational boundaries using standardized protocols. This model would promote innovation, market competition, and lower entry barriers for new players. However, it would require broad coordination around security, discovery, and interoperability, challenges that are technical, political, and economic in nature.

Conversely, dominant firms may restrict interactions between their agents and external agents, creating tightly controlled ecosystems. For instance, Meta launched free service agents in March 2025 for Facebook and Instagram business pages, but only for users inside its platforms. Similar restrictions are likely with Apple Intelligence, Google Assistant, Microsoft Copilot, and other branded AI ecosystems.

Walled gardens offer quality control and fraud prevention, but they also raise concerns about centralization, anti-competitive behavior, and innovation bottlenecks. A user might end up with separate agents for personal and professional contexts, unable to coordinate across them due to vendor lock-in. This fragmentation could limit the broader potential of AI-enabled economies.

What Will Drive Value in an Agentic Economy?

As AI agents reduce communication friction, several knock-on effects may fundamentally reshape market dynamics. Advertising, monetization, and even product design are expected to evolve.

In today’s economy, attention is scarce and advertising guides users toward transactions. But in an agent-mediated marketplace, discovery will hinge more on algorithmic matching between assistant and service agents than on banner ads or sponsored listings. Human feedback, especially high-quality, verifiable consumer preferences, will become the new currency of influence. This transition may signal a shift from an “attention economy” to a “preference economy”.

In addition, agent-based systems are likely to spur the rise of micro-transactions and dynamic product bundling. Agents can seamlessly coordinate payments on behalf of users, removing friction associated with small-value purchases. As a result, services that were traditionally locked behind subscriptions - music, news, or streaming - could be accessed via usage-based micro-fees. This transition may benefit content creators by enabling more granular monetization while offering consumers hyper-personalized, cost-efficient experiences.

Agents could also support real-time rebundling of content. A news assistant might compile custom articles from multiple sources, filtering out already-known information. Similarly, travel agents could assemble flights, lodging, and transportation based on user intent rather than pre-built packages. This model unlocks a market for componentized digital goods dynamically assembled through Retrieval-Augmented Generation (RAG) pipelines, provided that fair compensation for original content creators is integrated into the ecosystem.