Loose companion to a series on primitives for higher education. See The bundle is the challenge for the spine.

There's a pervasive conversation in higher education right now that follows the rough structure: the cost of learning has collapsed because of ubiquitous content, AI tutors are everywhere, anyone with motivation can master a subject from their phone, and so the four-year degree is obsolete and ought to be replaced with microcredentials, badges, or competency portfolios.

At worst, I find it dangerously misguided and likely to convince people who benefit most from college degrees to pursue a different path. The wage premium for a four-year degree hasn't gone away — College Board's most recent Education Pays data continues to show median earnings for bachelor's-degree holders running about 60% higher than for high school graduates, and the after-tax gap is large enough that the average bachelor's recipient recoups the cost of the degree by their mid-thirties.

At best, I find it unpersuasive, though not because I disagree about what's happened to the cost of learning. The reason is that the degree isn't doing the learning job. It's doing the recognition job, and recognition runs on different economics.

This is a companion piece to the main series I've been working on, that explores one particular topic critical to the work I'm doing on a day-to-day basis. It also connects to my frustration with the frame of "transformation" as the thing we're trying to do in high school and college. It explains why I think efforts that feed into the existing degree credential rather than around it are much likelier to expand access, lower cost, and have, yes, transformational impact. The argument leans on Ronald Coase, who worked it out in 1937, and applies to credentials almost without modification.

The lesson from Coase

Coase asked a question that sounds deceptively simple: why do firms exist at all? In a frictionless market, every task could be contracted out individually. Need accounting? Hire a freelancer. Need parts? Find a supplier. Need a marketing campaign? Negotiate one transaction at a time. Why do millions of people instead show up to the same building every day, working under one employer?

His answer was transaction costs. Every time you go to the open market, there are costs that don't appear on the price tag: finding the right party, evaluating their competence, negotiating terms, writing a contract, verifying quality, resolving disputes. When those costs accumulate to the point where they exceed the cost of standing relationships and internal coordination, it becomes cheaper to put people inside an organization and to rely on the organization for reputation and quality control. Firms exist because markets are expensive to use repeatedly.

Coase later won a Nobel for the idea, and economists have extended it in a lot of directions, but the core observation is what matters here. Institutions in large part are answers to specific transaction costs. If you change the underlying transaction costs, the institutions tend to change. If you don't, they generally don't.

The degree as a transaction-cost reducer

When an employer reviews a job application, they face an information problem of the kind Coase was describing. Is this candidate competent? Can they learn? Will they follow through on hard things? Can they operate inside a system?

A degree from an accredited college doesn't answer those questions perfectly, but it does answer them legibly and based on decades of track record. It compresses thirteen years of schooling, 2-4 years of college-level work, exposure to a discipline, completion under sustained pressure, and a baseline of cultural and institutional fluency into a single credential the labor market already understands. The employer doesn't have to evaluate it from scratch. They've evaluated thousands of degrees from this institution before, or from peer institutions, or from accredited institutions generally. The recognition cost has been amortized across decades and millions of hires.

That word legible is doing the same work it does in the rest of this series. The degree is an interface, in roughly the same sense that the shipping container is an interface or that TCP/IP is an interface. Its value is that everyone mostly agrees on what it means, not that it's theoretically optimal.

This is the fuller version of an argument from the post on the bundle. The bundle persists in part because institutions are sticky, but the larger reason is that the bundle's central output, the degree, has solved a real and durable transaction-cost problem in the labor market. Reforms that try to replace the bundle without replacing what the degree does in the labor market won't move learners through the actual market they're trying to enter. They'll move learners into a parallel credential system the labor market hasn't agreed to recognize, leaving the learners who think they've done what they need with a credential employers don't read.

Why microcredentials struggle

Think about the substitution the microcredential argument proposes. The bachelor's degree is replaced by dozens of micro-certificates from different platforms, AI tutors, bootcamps, and assessment providers. The employer's hiring problem now has a different shape. What does this badge mean? Who issued it? How rigorous was the underlying assessment? How does it compare to a similar-sounding badge from a different provider? Is any of it auditable?

Each of those is a transaction cost. Stacked together, they're a bigger transaction cost than the one the degree was solving. The substitution moves friction from the supply side, where producing a credential used to be expensive, to the demand side, where evaluating a credential is now expensive across every hiring decision. Employers will demand legibility one way or another. Without a shared signal, they'll either rebuild it informally — referrals, prestige proxies, in-house assessments — or restrict their hiring to candidates who already carry one (which is to say, candidates with degrees).

(Note that a version of this played out with standardized testing for selective college admissions. Advocates of doing away with standardized tests hoped they would drive more access and equity. What the policy change probably did was just shift decision-making to rely on a set of proxies that are arguably even less available to low-income or low-SES learners.)

There's a second problem layered on top. Currencies, languages, file formats, and credentials all share a property: their value to any one party rises with the number of other parties who already accept them. Network effects make incumbent standards hard to displace, even by something theoretically better. The dollar wins as a medium of exchange because everyone accepts it. The QWERTY keyboard wins because everyone learned to type on it. The shipping container wins because the ports are built for it. Proposing a microcredential as a substitute for the degree is structurally similar to proposing a new currency: the thing you're replacing has a large adoption advantage that's hard to overcome no matter how clever the alternative. This hurdle is steeper for microcredentials because employers and others who refer to degrees are likely to require years or decades of evidence of the quality of the microcredentials before they will even countenance a switch, leaving learners with those microcredentials in the lurch while the evidence base is developed.

How this connects to the rest of the series

There are two specific connections worth naming to the past ideas in the series and the ones to come.

First, this post details why the articulation layer matters so much. Articulation — the process by which a passed exam becomes credit on a transcript at a specific institution — is a layer where things break in the current system, and it's the layer Modern States only partially owns. The Coase frame describes why it's so important: it's where a primitive (a course, an exam) gets translated into an existing legible signal. In most cases, it works for CLEP exams because institutions recognize the knowledge tested on CLEP and the exams themselves have a track record showing that students who pass them generally do as well as or better than peers who took the equivalent introductory course on campus when both groups continue into the subsequent class. Upstream primitives can be excellent and still fail to benefit the learner if articulation fails, because without articulation the primitive doesn't compose into anything the ultimate audience has agreed to recognize. Articulation is the recognition interface.

Second, this is the demand-side complement to a question I'll explore later in the series. The case for treating primitives as ubiquitous public infrastructure is a supply-side claim. Once built, digital primitives can be reused and distributed at near-zero marginal cost, while improvement can be funded as a shared infrastructure investment rather than captured inside closed platforms. Coase is the demand-side analogue. The reason to compose primitives into the existing degree credential rather than around it is that the degree is the legible interface the labor market already runs on, and the cost of replacing it is enormous.

The two arguments together form a large component of our strategy at Modern States. The aim is to make the existing legible credential dramatically cheaper to acquire by composing free primitives that articulate into degree credit. CLEP is the specific assessment the current architecture leans on, but the argument doesn't depend on CLEP particularly; it depends on the principle that the assessment endpoint must compose into the labor market's existing legible signal. Anything else requires rebuilding the recognition layer from scratch, and the recognition layer took a hundred years to build.

What this argument doesn't claim

I'm making an argument here based on how I think the world actually works. I think a world where learning could be validated in a really wide range of authentic and non-standardized ways would be great, but I don't think we mostly live in that world. Until we do, it's worth being honest about that point, particularly for learners who understand a college degree substantially as a bridge to a more economically secure life.

There are existing contexts where microcredentials can be valuable that prove the broader point. They can do real work inside contexts where the issuer's reputation is established and the use case is bounded — internal training, vendor-specific certifications, narrow technical badges valued by a particular sector. They can add value for jobs or sectors where standardized credentials do not already exist, where employers already face high transaction costs, or where the knowledge and skill set is too new to have an alternative evidence base.

Equilibria of this kind can shift, slowly. AI may change what employers actually need to assess. New institutions may earn the right to produce a recognized credential of their own; Khan Academy, TED, and ETS are trying, as one example. The bundle may eventually unbundle from the inside, as employers gain better and cheaper tools for evaluating candidates directly. None of that is impossible, and some of it is probably already happening at the margin.

But at least for most people, for the foreseeable future, what they can't easily do is what the degree does: compress a candidate's underlying competence, persistence, and trainability into a signal that any employer in any sector recognizes without further evaluation.

Sources & references

On Coase and signaling

• Ronald H. Coase, "The Nature of the Firm," Economica 4, no. 16 (1937): 386–405. The original paper. Short, readable, and still does most of the work for arguments of this kind.
• Bryan Caplan, The Case Against Education: Why the Education System Is a Waste of Time and Money (Princeton University Press, 2018). A well-developed argument about the signaling component of higher education.

On test-optional admissions and proxy substitution

• David Deming, "The Worst Way to Do College Admissions," The Atlantic, March 2024. Argues that when colleges evaluate applicants without SAT or ACT scores, the remaining measures (recommendation letters, extracurriculars, essays) are even more prone to socioeconomic bias than test scores are.

On the persistence of the wage premium

• College Board, "Education Pays 2026" (Trends in Higher Education Series, 2026). Most recent edition of the long-running College Board series. Documents the continuing earnings, employment, and after-tax gap between bachelor's-degree holders and high school graduates.

On CLEP outcomes

• Jing Feng and Jeff Wyatt, "The Validity of CLEP Scores for Course Placement Decisions" (College Board Research, May 2025). Across 17 CLEP exams in five subject areas, students who earned CLEP credit performed the same or better than peers who took the equivalent introductory course on campus, with subsequent-course GPA differences favoring CLEP students by 0.11 to 0.82 points.

On standards and network effects

• Marc Levinson, The Box: How the Shipping Container Made the World Smaller and the World Economy Bigger (Princeton University Press, 2006). The central example of how standards win on adoption rather than theoretical merit.

Loose companion to a series on primitives for higher education. Check out the archive for prior posts.