Coasean Compression

On incomplete contracting at scale.

Building on Coase’s work on transaction costs, this essay looks at how markets at scale lose sight of our deeper wants, as some kinds of value resist being written into contracts, and markets coordinate around what can be (absent norms and reputation to fill the gap). It’s the first of three on the problem and what to do about it.

Edward Hopper, Nighthawks, 1942.

Introduction

Markets are astonishingly good at giving people what they want. At least, for certain kinds of wants. You can order a computer with more processing power than a supercomputer from the 90s and have it delivered tomorrow, or book a flight across the world in seconds.

But when it comes to some of the things people want most — love, friendship, belonging, meaning — the trend seems to run in the opposite direction.

Social media promised connection, yet the share of Americans with no close friends has quintupled since 1990.¹ Dating apps promised partnership, yet nearly half of Americans under 30 are single, roughly double the share three decades ago.² Many forces drive these trends of course, but the market options pitched as solutions clearly aren’t delivering.

So why is it that markets coordinate so effectively for things like laptops and logistics, yet struggle with many of our deeper wants?

This is the first of three essays exploring that question. Here, I introduce one reason markets lose sight of certain kinds of value at scale. The second essay examines why these losses don’t self-correct, and why this matters for labor displacement. The third, what kinds of market upgrades might help address this and better tie the economy to human flourishing.

Buying and Compression

When you buy something in the market you’re agreeing to a set of terms. Economists call this a contract: a specification of what’s to be delivered, at what price, under what conditions. When you buy a MacBook, for example, Apple promises you a machine with a certain processor, memory, display, and battery life, and you pay the given price. If you’re buying the MacBook to do software engineering, those specs track what you care about pretty well: a faster chip means your code compiles faster, more memory means more tabs and processes, a better battery means more hours at the café.

In such cases, where contracts track what you actually care about, markets work remarkably well. Every year, competition makes the MacBook better, faster and cheaper. Since what’s promised is legible and clear to both buyer and seller, you can return it if it’s not performing as intended. This is the case for electronics, appliances, raw materials, flights, package delivery, and so on.³ But many goods aren’t like that.⁴ Especially for our deeper needs, the contracts are incomplete at best.

Consider dating. From a dating service, you may want the conditions for finding a long-term partner, someone that clicks with you in a certain way. What you buy on a dating app is swipes. You’re left hoping these will convert to your underlying wants, and if they don’t (which is often the case), you can’t go back and demand a refund.

You could imagine a contract that tried to track what you’re really after more closely: “pay if a small number of genuine connections form within some time window, pay more if one of them turns into a lasting relationship, pay nothing if you got nothing out of it”. But this contract becomes untenable: the thing being promised (genuine connection, a lasting relationship) is hard to specify in advance (what is meant by genuine connection?), hard to independently verify (did genuine connection occur?) and risky for the supplier (whether “genuine connection” occurs hinges on factors outside of their control). More complete contracts exist at the high end — elite matchmakers charge success fees tied to marriage, for example. But that works because marriage is a legal event you can verify, it doesn’t work for qualitative outcomes.⁵

When there is a big skew between what’s being contracted around and what’s actually valuable to the customer, the business model no longer has to be tightly coupled to delivering user value. In the case of most dating apps, providers are incentivized to avoid delivering on what the customer actually wants as that would mean a churned user.

When Value Depends on Other People

A dating platform is arranging encounters between two people. For many goods, your experience depends on a whole group, which makes things even trickier.

Consider a neighborhood pub. Technically, customers are buying beer, but they come for the sense of familiarity and chance encounters; in other words, the “vibe.” But this vibe is an emergent property from who shows up and how they show up. You can’t swap out the crowd for random strangers and put a new hire behind the bar without killing the vibe. The contract captures only the private good (the beer). What people actually value is what economists would call a “positive externality”.

You could imagine a contract that prices in this externality: “pay more if the pub maintains a stable community such that familiar faces appear on most visits and you feel like a participant rather than a consumer, pay less if it turns into an anonymous throughput machine”.

But similar to the dating contract, the outcome is difficult to specify (what counts as a good vibe?), hard to verify (did it actually happen?), risky to promise (the pub doesn’t directly control the vibe), and additionally, dependent on the participation of others (who else walks through the door and how they show up).

The problem is not that these positive externalities exist, but that when they go unpriced, they’re foregone by optimization pressures from the market. At small scales, markets still work well because normative infrastructure (reputation and social norms) make up for the incompleteness of the contract. Customers who talk too loudly get sneered at by others until they get the hint. A bar owner who let the vibe decay has his personal reputation on the line.⁶ But a franchise chain with rotating staff serving anonymous customers across a thousand locations cannot be constrained by this kind of normative infrastructure. No one knows anyone else well enough to enforce norms and no one’s personal reputation is at stake when everyone’s an interchangeable customer or an interchangeable employee.⁷ And since the franchise bar has better unit economics, the franchise chain that optimizes for the explicit contract is poised to outcompete the neighborhood bar that optimizes for what people actually want.⁸

Costs and Compression

In 1937, the economist Ronald Coase asked a simple question: if markets are so efficient, why do firms exist? His answer, which later won him a Nobel Prize, was that using markets come at a cost. Finding trading partners, negotiating terms, and enforcing agreements all impose costs, and when those costs rise, coordination moves inside firms or other institutions.

But there is another possibility, which is what this essay is about: rather than moving activity out of markets, you can replace what’s being contracted around with a proxy; you can think of the contract as having been “compressed”. Maintaining a contract for genuine connection becomes extremely expensive, but maintaining a contract for swipes works perfectly well in terms of transaction costs. Platforms can then market “connection,” “community,” or “love” without being meaningfully accountable for whether those things materialize, whatever they may mean to the consumers seeking them.

I’ve hinted at the frictions that drive this compression in the examples above. They fall into roughly four categories:⁹

Specification. With a MacBook, the translation from intention to specification is straightforward. With a sense of belonging or a life partner, it is not. The work of translating deep wants into deliverable terms might require introspection to identify what it is you’re really seeking, and finding a way to write it out in language precise enough to be adjudicated.

Verification. Even if a deep want could be specified, verifying that it has been satisfied is hard. You can benchmark a MacBook. You cannot easily benchmark whether a relationship worked or whether a festival delivered the sense of serendipity you wanted out of it. These outcomes are often subjective, emergent, and only legible over long time horizons. Besides, evaluating a friendship can make it feel transactional, depriving it of its meaningfulness.¹⁰

Risk. Whether a MacBook delivers as promised is largely in Apple’s hands. Whether a date leads to genuine connection is to a large extent outside the immediate control of Tinder. Richer specifications expose suppliers to more risk from factors they don’t control: timing, mood, other people’s choices, contingency. Promising access is much safer than promising outcomes.

Configuration. A MacBook is a product sold to you as an individual. Events, parties, festivals, concerts, intellectual salons; their value to you as a participant depends on who else participates and how they show up. Trying to capture this complexity in a contract is very difficult, so most contracts are bilateral.¹¹

---

This is what I mean by Coasean Compression. When what people want is easy to specify, easy to verify, safe to promise, and independent of who else shows up, markets see clearly and coordinate well. When it isn’t, markets go partially blind, coordinating through proxies (at least at scale, where normative infrastructure is absent).

Hence we see a proliferation of dating apps and AI companions while becoming increasingly more lonely and isolated.

Two questions come up naturally here: if companies are only delivering thin proxies, why doesn’t competition eventually respond to this latent demand and produce the real thing? And if markets can’t handle dating or friendship or adventure, why not just find those things in real-world communities outside markets?

I address these in the next essay. The short answer is that the compressed version gets cheaper every year making the “fuller version” relatively more expensive, and additionally the compressed version can erode the conditions the non-market spaces you’d presumably exit into depends on.

In the third post, I’ll outline some ideas for how to upgrade markets to make sure what we actually care about becomes legible to them. It’s increasingly important, as a hope from economists is that human labor remains scarce due to our capacity for meaningful relational work, which is exactly the kind of “fluff” compressed contracts get rid of. I don’t think this is an intractable problem, especially not with AI. Markets have been reimagined many times in the past and it’s time to do it again.

Thanks to Joe Edelman, Merlin Stein, Maximilian Kroner Dale, Alex Chalmers, Scott Moore and Tobias Werner for comments.

1. In 1990, 3% of Americans reported having no close friends; by 2024, 17%. Cox and Pressler, “Disconnected,” Survey Center on American Life (2024). The U.S. Surgeon General declared loneliness a public health epidemic in 2023.

2. Pew Research Center (2023), “Key findings about online dating in the U.S.” About 47% of U.S. adults under 30 are single (not married, cohabiting, or in a committed relationship); among men under 30 the figure is approximately 63%. The share of 25–29-year-olds who are married fell from approximately 50% in the early 1990s to about 29% by the early 2020s (U.S. Census Bureau, Current Population Survey).

3. Economists distinguishes search goods (quality observable before purchase) from experience goods (observable only after, like a concert) and credence goods (unverifiable even after, like supplements). The goods discussed here are either experience goods or credence goods. Frost et al. (2008) uses this taxonomy to discuss dating.

4. On incomplete contracts, see Grossman & Hart (1986) and Hart & Moore (1990).

5. Holmström and Milgrom (1991): when tasks differ in measurability, strengthening incentives on the measurable dimension diverts effort from the unmeasurable one. See also Baker, Gibbons, and Murphy (1994).

6. Hadfield-Menell and Hadfield (2018) argue that human contracts work despite incompleteness because agents draw on external normative structure (culture, reputation, social sanctions) to fill gaps. The claim that this infrastructure erodes with scale is mine. At franchise scale, what emerges is a form of moral hazard: the operator is insulated from the reputational consequences that would discipline a local owner, and so has little incentive to maintain the uncontracted dimensions of quality.

7. Hirschman, Exit, Voice, and Loyalty (1970). At small scale, exit and voice are legible signals that keep providers accountable. At franchise scale, exit is invisible and voice has no receiver. Simmel (1903) identifies a related mechanism: metropolitan anonymity produces emotional withdrawal — when the room is full of strangers, people stop treating it as a space they’re responsible for.

8. Digital reputation systems (Airbnb, Yelp) are attempts to scale normative infrastructure, but they compress too: they work for individually-ratable dimensions (cleanliness, listing accuracy) and break down for emergent or configurational value. The drift toward a homogeneous Airbnb aesthetic could be seen as hosts optimizing for rated dimensions at the expense of unrated ones.

9. The four frictions are not transaction costs in the strict Coasian sense (search, bargaining, enforcement) but frictions in translating value into contractible form. Verification draws on incomplete contracts (Hart; Grossman); Risk on principal-agent theory (Holmström); Specification is related to but distinct from Stigler’s (1961) search costs (Stigler assumes you know what you want; the friction here is articulating inchoate wants); Configuration connects to network externalities and relational goods (Katz & Shapiro; Gui).

10. Sandel, What Money Can’t Buy: The Moral Limits of Markets (2012), argues that some goods are degraded by being brought into market logic: pricing a friendship changes what it is.

11. Market design (Roth, 2002; Milgrom, 2004) takes preferences as given and optimizes allocation for well-defined goods. The problem here is upstream: what happens when value cannot be articulated as a preference, or depends on group composition.