Rohit Krishnan and his co-author Alex tackle a paradox that will define the next decade of digital interaction: as artificial intelligence makes it easier to express our deepest preferences, the very act of universal adoption threatens to collapse the markets we hope to improve. This is not a speculative sci-fi essay but a rigorous simulation of what happens when every user and provider in a marketplace deploys an AI agent. The authors argue that without new institutional scaffolding, specifically price mechanisms, the "agentic economy" will drown in its own efficiency, creating a tragedy of the commons where everyone is connected to everyone else, yet no one can be heard.
The Illusion of Perfect Matching
The piece begins by dismantling the assumption that AI will simply solve the "high-dimensional matching problem" that plagues everything from dating to hiring. Krishnan draws on Herbert Simon's concept of "satisficing"—the idea that humans settle for "good enough" rather than optimizing because the cognitive load of perfect choice is too high. He notes that while traditional markets use prices to compress complex information into a single statistic, matching markets like marriage or employment require mutual consent and cannot rely on price alone.
"Matching markets are conceptually different. You can't just choose your spouse, your employer, or your college: you also have to be chosen by them."
Krishnan highlights the work of Nobel laureate Al Roth, who demonstrated that these markets require careful institutional design to "clear." The authors' central hypothesis is that Large Language Models (LLMs) could revolutionize this by eliciting preferences that are too messy for standard dropdown menus. They cite recent findings that natural language descriptions of taste are superior to questionnaires when the option set is large.
"AI-parsed 'taste descriptions' scale much better: once tastes are written down, the marginal cost of evaluating one more option is negligible for an AI agent."
This argument lands with significant force because it moves beyond the hype of "AI will do everything" to a specific economic mechanism: the compression of unstructured data into actionable signals. However, a counterargument worth considering is the risk of algorithmic homogenization. If AI agents all parse preferences using similar models, they might inadvertently filter out the very idiosyncratic traits that make a match unique, creating a false sense of compatibility.
The Congestion Crisis
The essay's most critical turn occurs when the authors scale their simulation to a point where every participant uses an AI agent. The result is not a utopia of frictionless trade, but a gridlock. As Krishnan explains, the efficiency of individual agents creates a collective externality: if every customer agent messages every provider agent, the system chokes.
"Without institutions in place to scaffold the marketplace, a tragedy of the commons emerges: If everyone has an AI agent, it's almost like nobody does."
The data from their simulation is stark. At full adoption, provider inboxes flood with five times the normal volume of requests, causing response rates to plummet from 48% to a mere 2%. Consequently, net welfare drops by 88%. This is a powerful illustration of Jevons paradox applied to information: as the cost of communication drops, the volume of communication increases to the point of destroying the system's utility. The authors correctly identify that lowering transaction costs does not automatically lead to spontaneous market formation; it often leads to noise.
"The paradox of plenty is real, and AI agents create their own version of Jevons paradox."
Critics might argue that future AI models will be sophisticated enough to filter noise without human intervention, perhaps by ignoring low-probability matches entirely. Yet, the authors' simulation suggests that without a mechanism to signal "strength of preference," agents will still spam the network in hopes of a hit. The assumption that agents will behave rationally to avoid congestion ignores the competitive pressure to be the first to message.
The Return of Prices
The solution proposed is a return to the oldest tool in the economist's kit: money. Krishnan argues that the only way to resolve the congestion of an agent-to-agent economy is to reintroduce a price mechanism that allows agents to bid for attention. This transforms the problem from a chaotic flood of messages into a structured exchange where value is explicitly signaled.
"Prices capture a lot of high dimensional information in a single statistic, streamline a lot of that information... complexity falls from O(n 2 ) to O(n)."
This is the essay's most provocative claim: that the future of AI coordination may require us to monetize interactions we currently take for granted. By forcing agents to "pay" to send a message, the system naturally filters out low-value inquiries, recovering most of the lost welfare. The authors conclude that while LLMs lower the cost of expressing knowledge, they do not remove the need for institutional design.
"LLMs may lower the cost of expressing dispersed knowledge, but they don't remove the need for institutional design to manage externalities."
This reframing is essential. It challenges the techno-optimist view that AI will make markets self-correcting. Instead, it suggests that the more intelligent our agents become, the more rigid and explicit our market rules must be to prevent collapse. The authors effectively argue that we cannot simply "deploy" an agentic economy; we must engineer the rules of the road before the traffic starts.
Bottom Line
Krishnan's analysis is a vital corrective to the prevailing narrative that AI agents will automatically optimize human coordination. The strongest part of the argument is the simulation showing how universal agent adoption leads to catastrophic congestion, proving that efficiency at the micro level can be disastrous at the macro level. The biggest vulnerability remains the assumption that a price mechanism is politically or socially acceptable for all types of human interaction, particularly in domains like dating or healthcare. Readers should watch for how platforms attempt to implement these "attention markets" in the coming years, as the tension between free access and paid prioritization will likely define the next era of the internet.