The Friction Merchants

What happens to intermediaries when the friction they solved dissolves

#

TAM-RWR.2-01 · The Reshaped World, Arc 2: The Invisible Ledger · The Approximate Mind

Caroline has given this explanation hundreds of times, and she gives it again now, at a dinner table in a restaurant where the bill will be split via a tap on a phone. She explains how Visa actually works. The authorization request traveling from the merchant’s terminal to the acquiring bank to the card network to the issuing bank and back in under a hundred milliseconds. The settlement cycle. The interchange fees nested inside fees nested inside fees. The chargeback architecture. The fraud detection layer. The risk models running beneath every transaction, constantly, updating in real time.

The people listening have used the product thousands of times.

None of them have ever thought about it.

There is a glass jar on her desk at the office, accumulated from twenty years of international travel: foreign coins, a hundred of them maybe, from currencies she can half-name. Lira, kronor, forint, dirham, won. She has never counted them. She does not know their total value. She keeps them because she likes the weight of the jar when she picks it up, the sense that she is holding something real from many different places, the accumulated proof of having been in the world.

The Toll Booth at Origin

#

Every financial intermediary was built to solve a genuine problem.

Not a convenient problem, not a manufactured problem, a real one: the problem of conducting economic exchange with strangers across distances in a world without trust infrastructure. You could not know whether the merchant in the next town was honest. The bank intermediated trust. You could not predict whether your house would burn down or your ship would sink and whether the loss would be survivable. Insurance pooled the unpredictable across enough people to make it manageable. You could not transfer value across continents without a physical mechanism for doing so. The correspondent banking network made the mechanism.

Each intermediary charged a fee that was, at origin, the price of the problem it was solving. The fee was not extraction. It was payment for a service the market had not found another way to provide. The banker earned the spread because assessing creditworthiness required expertise and relationships the borrower did not have. The insurance broker earned the commission because navigating the underwriting market required knowledge the policyholder could not efficiently acquire. The payments network earned the interchange because building and maintaining the infrastructure of trust across millions of merchant relationships was expensive and difficult.

This is worth sitting with. The toll booth economy did not begin as rent extraction. It began as genuine value creation. The intermediaries solved real problems. They built real infrastructure. They employed real expertise. The problem is not that they charged for this. The problem is what happened next.

The Persistence Mechanism

#

Toll booths outlive their function.

This is not a metaphor. It is an observable pattern in market structure. The highway toll booth built to fund the road’s construction collects tolls decades after the road is paid for, because removing the toll booth requires a political decision that the toll authority has no incentive to make. The financial intermediary built to solve the trust problem continues charging for trust assessment after AI has made trust assessment cheap, because removing the chokepoint requires a competitive disruption that the intermediary’s network effects make difficult to achieve.

The persistence mechanism has three components. First, switching costs: financial relationships are embedded in systems, contracts, habits, and institutional processes that make changing them expensive even when the alternative is cheaper. Second, regulatory capture: intermediaries that have operated for decades have shaped the regulatory environment they operate within, making new entrants comply with legacy rules designed for the problem the intermediary no longer solves. Third, information asymmetry: the intermediary still controls the data infrastructure that would reveal how much of its fee is now extraction rather than service.

Caroline’s industry understood all three. She did not think of them as rent protection. She thought of them as competitive moats.

The distinction matters less than it sounds.

Three Dissolving Frictions

#

Payments processing is the clearest case because the dissolution is most advanced.

The original problem: merchants could not trust that payment would arrive, and consumers could not trust that merchants would protect their financial information. The card network solved both: guaranteeing settlement to the merchant and providing dispute resolution to the consumer. Interchange fees, on the order of 2 percent of every transaction, were the price of this guarantee.

AI dissolves the problem in two directions. On the fraud side, real-time transaction monitoring is now orders of magnitude more accurate than the fraud detection built into the original interchange pricing, meaning the risk the fee was priced to cover has declined dramatically while the fee has remained constant. On the settlement side, distributed ledger protocols and central bank digital currencies are making guaranteed settlement possible without the card network’s infrastructure, which means the infrastructure’s value proposition is migrating from genuine service to regulatory incumbency.

The toll booth is collecting the same toll for a bridge whose construction loan was paid off a generation ago.

Insurance brokerage is a slower dissolution, but the direction is identical. The original problem was information asymmetry: the policyholder could not efficiently navigate an opaque underwriting market, and the broker’s expertise reduced the cost of finding appropriate coverage. AI eliminates the asymmetry. The policyholder’s AI agent can now analyze coverage terms across carriers with a thoroughness no human broker achieves, identify the relevant exclusions, model the premium trajectory, and present options ranked by total cost of risk. The broker’s information advantage, which was real when the information was difficult to synthesize, is dissolving as synthesis becomes trivial.

Mortgage origination is the most socially consequential case. The original problem was genuine: assessing the creditworthiness of a borrower for a thirty-year obligation required integrating income, assets, employment history, property value, and local market conditions in ways that required both expertise and access to data. Loan officers performed this function. They also, systematically and demonstrably, introduced human bias into the assessment: racial steering, discriminatory pricing, preferential treatment of familiar profiles.

AI credit assessment is, on the averages, more accurate and less biased than the human alternatives. The dissolution of the friction is real. The chokepoint has not dissolved with it.

What the Consumer’s Agent Sees

#

The technology that makes friction assessment cheap also makes the toll booth visible.

A consumer using an AI agent to compare mortgage offers can now see, in a format no previous generation could access, the effective interest rate spread above the risk-free rate, the origination fee decomposed into its components, the comparison across lenders adjusting for terms, and the historical pattern of rates for borrowers with identical risk profiles. The opacity that made the toll booth extraction sustainable is dissolving.

I wonder whether consumers will act on what their agents can now show them, or whether the behavioral inertia of financial relationships, the tendency to use the same bank, the same broker, the same insurer because switching feels complicated regardless of its actual cost, is strong enough that people will continue paying the toll even after they can see it clearly and the friction it was built to solve is gone.

The evidence from markets where AI comparison tools exist suggests the answer is: some will, and the some is not uniformly distributed. The consumer with the AI agent who understands how to interpret its outputs, who has the financial literacy to act on the comparison, who has the time and cognitive bandwidth to execute the switch, will route around the toll. The consumer who does not have those things will continue paying.

The toll booth doesn’t disappear when AI makes it visible. It becomes a stratification mechanism.

The friction merchants face a choice that most of them are not yet naming as a choice: become actually useful at what the AI agent cannot yet do well, or defend the chokepoint through regulatory means until the regulatory protection fails. The first path requires reinvention. The second path requires only incumbency, which is the easier path, and therefore the path most intermediaries are on.

The Chokepoint Economy

#

There is a version of this story that ends optimistically: AI dissolves the friction, the toll booth revenues migrate to productive uses, the financial system becomes more efficient, and the savings flow to consumers. This version is accurate about the technology and naive about the political economy.

Financial intermediaries do not dissolve gracefully. They adapt. The payments networks adapting to the threat of protocol-based settlement are buying the protocol companies, integrating them, and capturing the new infrastructure behind the old regulatory perimeter. The mortgage industry adapting to AI origination is lobbying for credit assessment standards that require human review, which requires the human reviewer, which maintains the origination fee. The insurance industry is using AI to improve its own underwriting while at the same time opposing the consumer-facing AI tools that would make the improvement visible as extraction.

The chokepoint economy does not eliminate itself. It migrates. The specific form of rent extraction changes. The rent extraction continues.

What the friction merchants are actually selling, in the transition, is regulatory position. Not the solution to the original problem, not the infrastructure that once made the solution costly, but the accumulated relationships with the regulatory bodies that certify the legitimacy of their role. The moat is now political, not technical. The technical moat has been dissolved by AI. The political moat is stronger than it has ever been, because the incumbents have more resources than they have ever had to maintain it.

After the Explanation

#

Caroline finishes explaining how the system works. The table is quiet for a moment.

Someone asks: so it’s basically a tax on every transaction, and the tax goes to a company in the middle that doesn’t actually do anything except maintain the infrastructure of the relationship?

She does not say yes. She explains the fraud protection. She explains the dispute resolution. She explains the network effects that make the system valuable to merchants precisely because consumers trust it. She explains these things because they are true.

She also knows that the consumer’s AI agent can now provide equivalent fraud protection through different means, that dispute resolution is increasingly automated, and that the network effect that once required a centralized intermediary can now be achieved through distributed protocols. She knows that the honest version of the answer to the dinner table question is closer to yes than she is comfortable saying in public.

The jar on her desk has, she estimates, about three pounds of coins in it. She has never calculated their value. She keeps them because they are real in a way that the numbers she works with are not, solid and specific, each one a particular place at a particular moment of her career.

She puts her card down for the bill. The authorization travels from the restaurant’s terminal to the acquiring bank to the card network to the issuing bank and back in under a hundred milliseconds.

The infrastructure is real. The question is what it’s for now.

She picks the jar up on her way out of the office sometimes, just to feel the weight. Their value is uncertain and their weight is not.

References

#

Financial Intermediation and Its Origins

Gorton, Gary. Misunderstanding Financial Crises: Why We Don’t See Them Coming. Oxford University Press, 2012.

Mian, Atif, and Amir Sufi. House of Debt: How They (and You) Caused the Great Recession, and How We Can Prevent It from Happening Again. University of Chicago Press, 2014.

Minsky, Hyman P. Stabilizing an Unstable Economy. McGraw-Hill, 2008.

Payments and the Toll Booth Economy

Dunn, Evelyn, and Leora Klapper. “Digital Financial Inclusion: Current Policy and Practice.” World Bank Economic Review, vol. 34, no. 1, 2020, pp. 1–19.

Philippon, Thomas. The Great Reversal: How America Gave Up on Free Markets. Harvard University Press, 2019.

Stigler, George J. “The Theory of Economic Regulation.” Bell Journal of Economics and Management Science, vol. 2, no. 1, 1971, pp. 3–21.

AI and Financial Intermediation

Berg, Tobias, et al. “On the Rise of FinTechs: Credit Scoring Using Digital Footprints.” Review of Financial Studies, vol. 33, no. 7, 2020, pp. 2845–2897.

Fuster, Andreas, et al. “Predictably Unequal? The Effects of Machine Learning on Credit Markets.” Journal of Finance, vol. 77, no. 1, 2022, pp. 5–47.

Navaretti, Giorgio Barba, et al. “Fintech and Banking: Friends or Foes?” European Economy – Banks, Regulation, and the Real Sector, vol. 2, 2018, pp. 9–30.

Regulatory Capture and Incumbency

Kwak, James. “Cultural Capture and the Financial Crisis.” Preventing Regulatory Capture: Special Interest Influence and How to Limit It, edited by Daniel Carpenter and David A. Moss, Cambridge University Press, 2013, pp. 71–98.

Zingales, Luigi. A Capitalism for the People: Recapturing the Lost Genius of American Prosperity. Basic Books, 2012.