Could agent-to-agent communication reshape market power?

Rothschild, Mobius, Hofman, Dillon, Goldstein, Immorlica, Jaffe, Lucier, Slivkins, and Vogel — researchers at Microsoft, with corresponding author David M. Rothschild — argue in this working paper that the "more profound economic impact" of generative AI agents is not the productivity gains from automating tasks consumers and businesses "had already been doing," but the reduction of "communication frictions between (and among) consumers and businesses." They write that consumers "have traditionally faced high communication costs when initiating relationships with businesses," illustrating with a consumer who hesitates to switch tax preparers because she would have to "explain her financial situation all over again." Lowering these costs, the authors argue, "expand[s] the landscape of options available to both consumers and businesses, increasing the efficiency of markets," with effects they expect to "reorganize markets, redistribute power, and catalyze the creation of new products and services."

The paper's mechanism rests on two preconditions: first, that consumers and businesses widely adopt assistant and service agents, which the authors say is "already underway"; second, that these agents be designed to "programmatically interact with each other to facilitate transactions" rather than merely simulate a human clicking through an unchanged interface. The authors argue today's agents fall short of this second condition. "Siloed service agents" like Amazon's Rufus or Expedia's Romie offer a chat interface into a single company's own data but expose no interface for other agents to call. "End-to-end agents" from OpenAI, Google, and Microsoft mostly rely on "computer use models that simulate a user pointing and clicking on existing (non-agentic) websites," which the paper says risks "adversarial relationships with businesses" that depend on advertising revenue and may resist agent traffic.

This sits alongside Will agents compete for attention just like users do?, which shares the premise that agent-mediated interaction displaces direct human interfaces as the site of competition; Rothschild and colleagues add the stronger claim that this shift lowers switching costs enough to reorganize market power, not just redirect which interface competes for attention. It also qualifies Why do consumers delay adopting agentic search?: where that paper finds platforms may have revenue reasons to resist agent-optimal search, this paper's account of why platforms might resist "agent-based access" — protecting ad revenue and control over presentation — supplies a concrete mechanism for that resistance.

As a working paper, not peer-reviewed, the argument is conceptual rather than empirical: it offers no measurement of how much communication frictions actually cost consumers or businesses today, nor evidence on how much an agentic economy would lower them, only the illustrative tax-preparer example and the assertion that adoption is "already underway." Whether friction reduction actually redistributes market power, as opposed to simply shifting which incumbents capture it, is a question the paper itself defers to its discussion of walled gardens versus an open web of agents — see Will agent economies develop as open webs or closed gardens?.