Updated: December 2, 2025
Cost of transactions drops - signaling cost rises. While some portion of the cost of transactions in AI-agent-dominated markets may drop, the transaction cost, which includes the signaling and counterparty risk cost, will inevitably increase.
Transaction costs are like the first law of thermodynamics. Once one cost drops and another rises, the balance of the total cost is maintained, though the distribution of cost changes. Since the distribution of costs changes, the inequality rises in favor of agents with more ample signaling resources.
Collapse of signaling. The rising signaling competition at the low cost of sending signals may possibly lead to the collapse of effort-based signaling. This effect may be smoother in decentralized markets due to higher specialization, but even if they do not necessarily mitigate the negative externalities of the reduced cost of transactions.
Introduction
With my ongoing involvement in the Digital Business Online Seminar, I am getting increasingly exposed to the working papers on AI economics. While my core expertise lies in the blockchain technology domain, I set a goal to develop a deeper understanding of the AI theme for potential research. At the end of the day, there are eternal mechanisms, like information asymmetry or transaction costs, that are at the foundation of every technological phenomenon once it hits the labor market and economic relationships.
One of my latest reads by Shahidi et al. (2025) raises an interesting topic of demand, supply, and market design in the presence of AI agents. Its key argument centers on the low transaction costs in AI-mediated markets.
In this article, I am exploring the (bold) idea that total transaction costs are relatively constant at the market level; i.e., if costs reduce in one place or process, they rise somewhere else. In this sense, transaction costs behave like the first law of thermodynamics: they transform from one scale or domain to another but never disappear fully. If AI agents dramatically reduce the costs of search and inference, the key question becomes: What kinds of costs increase instead?
What kinds of costs increase
One plausible answer to the previous question is that signaling and reputation costs rise. In an environment where almost anyone can do almost anything cheaply, the quality bar rises, and it becomes substantially harder to distinguish between good, bad, and truly terrible outputs. Expected standards go up, and the market becomes noisier. As a result, differentiation becomes more difficult and more expensive.
Citing one of Shahidi et al's illustrations as an example here: if AI agents end up consuming all advertising content, ad spending becomes a pure waste from the perspective of human consumers. Brands then stop investing in advertising, consumers no longer receive informative signals, and the signaling layer of the market risks collapsing.
One possible consequence of this situation is market fragmentation and increased distance between market segments. Segments become more internally homogeneous but more distant from one another. Reaching into adjacent or new segments becomes costly because of this “distance,” even if basic production and search are cheap.
Another externality of this shift toward higher signaling costs is the amplification of inequality. High-quality signaling requires substantial resources — time, capital, data, and trust in the brand. This can drive a redistribution of resources from the capable majority (newly equipped employees and smaller firms) toward the owning minority (large platforms and tech giants that can afford persistent, large-scale signaling and reputation-building).
How it relates to decentralization
The inequality argument does not straightforwardly apply to decentralized markets, but it does not contradict it either. In a decentralized market, there is often sharper specialization of tasks and clearer criteria for what counts as a “good” node. Nodes exert effort in a well-defined way on the supply side. However, on the demand side, if AI-driven agents eliminate much of the effort of search and evaluation, the question returns: how do we distinguish quality? In practice, this may occur through realized welfare, which is unequally distributed, as it benefits those who end up better off.
Conclusion
If we take seriously the idea that transaction costs in AI-intensive markets do not vanish but migrate, then transformative AI may be less about eliminating frictions and more about reshaping them. Lower search and inference costs can be offset by higher signaling and reputation costs, leading to market fragmentation and potentially greater inequality. Decentralized markets may mitigate some of these dynamics through sharper task specialization and transparent performance criteria on the supply side, but they do not fully resolve the demand-side problem of quality differentiation when human effort is automated away. While Shahidi et al. (2025) bring about several market design issues for AI agents — Sybil attacks, proof of humanity, fraud, and congestion — they all add the unaccounted overhead cost rooted in the loss of trust towards the markets and economic agents.
Shahidi, P., Rusak, G., Manning, B. S., Fradkin, A., & Horton, J. J. (2025). The Coasean singularity? Demand, supply, and market design with AI agents (Chap. 6). In A. K. Agrawal, A. Korinek, & E. Brynjolfsson (Eds.), The economics of transformative AI. University of Chicago Press. https://www.nber.org/books-and-chapters/economics-transformative-ai/coasean-singularity-demand-supply-and-market-design-ai-agents
Please cite this article as:
Petryk, M. (2025, November 10). Agentic AI: Do the Frictionless Markets Truly Exist? MariiaPetryk.com. https://www.mariiapetryk.com/blog/post-27