The Education Question
A structural analysis of education — what it was built to do, what it claims to do, and what it actually does.
Education is one of the few institutions that nearly everyone passes through, nearly everyone has opinions about, and nearly everyone misunderstands. The discourse around it tends to collapse into familiar grooves: progressives want more funding, conservatives want more discipline or more choice, libertarians want the state out entirely, and technologists want to replace the whole thing with apps. Each of these positions contains fragments of legitimate insight embedded in frameworks that obscure more than they reveal.
The purpose of this essay is not to advocate for a particular reform agenda. It is to examine the structural logic of education as it actually exists — the historical forces that shaped it, the economic incentives that sustain it, the ideological assumptions it encodes, and the ways it systematically differs from what it claims to be. The goal is analytical clarity, not policy prescription. Understanding what the machine does is a prerequisite for any honest conversation about what it should do.
1. The Industrial Legacy of Public Education
The Factory Model and Its Logic
The system of universal compulsory education that exists in most Western nations was not designed to cultivate the life of the mind. It was designed to produce a workforce. This is not a conspiracy theory or a revisionist reading — it is the explicit, documented intention of the people who built it.
The Prussian model of compulsory schooling, which became the template for American public education in the mid-19th century, was engineered around a specific set of industrial requirements: standardization of output, obedience to hierarchical authority, tolerance for repetitive tasks, adherence to externally imposed schedules, and the sorting of individuals into functional categories. Horace Mann, widely regarded as the father of American public education, traveled to Prussia in 1843 and returned with an enthusiastic report on a system that produced reliable, compliant citizens at scale. The design was not hidden. The bells that segment the school day into uniform periods mimic factory shift signals. The arrangement of students in rows facing a single authority figure mirrors the foreman-worker relationship. Age-based batching — grouping children by birth year rather than by ability, interest, or developmental stage — reflects the industrial preference for interchangeable units over individual variation.
This is not to say the system was malicious in intent. The 19th-century architects of public education faced a genuine problem: how to integrate massive waves of rural and immigrant populations into an industrializing economy that required basic literacy, numeracy, and socialization. The factory model was, in its context, a reasonable engineering solution to a real logistical challenge. It worked. Literacy rates climbed. A shared civic vocabulary emerged. Children who would otherwise have spent their lives in agricultural labor or urban destitution gained access to knowledge that had previously been the exclusive province of the wealthy. The democratization of basic education was one of the most consequential achievements of the modern era, and dismissing it as mere social control is as intellectually dishonest as pretending it was purely benevolent.
But the context that justified the design has largely evaporated. The industrial economy that needed standardized workers has given way to an information economy that ostensibly values creativity, adaptability, and self-direction — precisely the qualities the factory model was designed to suppress. The system persists not because it continues to serve its original purpose well, but because institutions, once established, develop their own inertia. They create constituencies — administrators, unions, textbook publishers, testing companies, real estate markets tied to school district ratings — whose interests become entangled with the system’s perpetuation regardless of its educational efficacy.
The Childcare Function
Any honest analysis of public schooling must reckon with a fact that is universally understood and almost never explicitly stated in policy debates: schools function as childcare infrastructure. For the majority of American families, the school day is what makes dual-income households possible. The summer learning loss that education researchers document every year is real, but the reason the school calendar still follows an agrarian schedule has less to do with pedagogy than with the political impossibility of extending the school year against the preferences of families, tourism industries, and the teachers who would have to staff it.
This childcare function is not incidental — it is load-bearing. When schools closed during the COVID-19 pandemic, the economic disruption was immediate and severe, falling disproportionately on women and lower-income families. The speed with which “reopen schools” became a bipartisan demand had relatively little to do with concern about learning loss and almost everything to do with the fact that the economy cannot function when parents of young children have no place to put them during working hours.
Acknowledging this does not diminish the educational mission of schools. It does, however, clarify the structural constraints under which any reform must operate. Proposals that would improve education but disrupt the childcare function — radically flexible scheduling, self-paced learning that doesn’t map onto a 7-to-3 institutional day, apprenticeship models that place teenagers in workplaces — face resistance not because they are pedagogically unsound but because they threaten the labor-market infrastructure that schools silently provide. The dual function creates a kind of structural conservatism: the system cannot change in ways that would require the simultaneous reinvention of how families manage the logistics of daily life.
The Myth of Curricular Neutrality
A subtler but more consequential feature of compulsory public education is the structural impossibility of curricular neutrality. The question is not whether schools transmit values — they inevitably do. The question is whose values, selected by whom, through what process, and with what degree of transparency.
Every curricular decision is a political act. The choice to teach American history as a narrative of expanding liberty rather than as a narrative of contested power is political. The choice to teach it as a narrative of systemic oppression rather than as a narrative of imperfect but genuine progress is equally political. The choice to present both narratives and let students decide sounds neutral but is itself a political commitment to a particular epistemological framework — liberal pluralism — that not all communities share. There is no view from nowhere. The textbook that seems merely factual is factual in a way that reflects the priorities, blind spots, and ideological commitments of the people who wrote it, the committees that approved it, and the political environment in which those committees operated.
This is not an argument against public education or against having a curriculum. It is an argument against the pretense that any curriculum is simply “teaching the facts” or “staying out of politics.” The facts themselves are selected from an infinite set of possible facts, and the selection criteria are never politically innocent. When Texas and California produce different history textbooks — as they reliably do — neither version is the neutral one. They are different political documents produced by different political cultures, each claiming objectivity while encoding particular assumptions about what matters, what counts as progress, and whose experiences are central to the national story.
The structural implication is significant: compulsory education necessarily involves the state compelling children to absorb a particular account of reality for twelve or thirteen years during the most formative period of cognitive development. The degree to which this is troubling depends on one’s political priors, but the fact of it should be acknowledged rather than obscured behind the language of neutral expertise. The progressive who insists that inclusive curricula are simply “accurate” and the conservative who insists that traditional curricula are simply “patriotic” are making the same rhetorical move — disguising a normative choice as a descriptive one. Analytical honesty requires recognizing that both are engaged in the same activity: using the coercive apparatus of compulsory education to transmit a preferred vision of the world to other people’s children.
None of this means the enterprise is illegitimate. A society that educates its children must teach them something, and that something will always reflect choices. But the choices should be made — and contested — openly, rather than laundered through a false claim of objectivity that serves primarily to insulate the curriculum from democratic scrutiny.
2. The University as Social Sorting Engine
What Universities Actually Do
Higher education is discussed as though it were a single institution with a single purpose. It is not. The modern university is an unstable bundle of at least six distinct functions that have been lashed together for historical and economic reasons, not because they logically cohere: credentialing, social sorting, professional gatekeeping, research, cultural reproduction, and networking. Understanding why the system resists reform requires understanding that any proposed change threatens at least one of these functions, and each function has its own constituency prepared to defend it.
Credentialing is the most visible function and the one most often confused with education itself. A bachelor’s degree signals to employers that the holder can show up, follow instructions, meet deadlines, and tolerate boredom for four consecutive years — qualities that correlate with workplace reliability regardless of whether the holder learned anything of substance. The degree is a filter, not a measure of knowledge. This is why employers routinely require degrees for positions that do not require any of the knowledge the degree ostensibly certifies. The HR department is not confused about what a communications major learned. It is using the degree as a low-cost screening device for conscientiousness and class background.
Social sorting is the function that dare not speak its name. Elite universities admit a class that is, by design, a curated social environment: the children of the wealthy, the conspicuously talented, and a carefully managed number of lower-income students whose presence provides both moral legitimacy and evidence that the system is meritocratic. The sorting happens before any education occurs. The day an eighteen-year-old is admitted to Harvard, their life trajectory changes — not because of anything they will learn in a Cambridge lecture hall, but because they have been stamped with a marker that will follow them for the rest of their professional life. The education is, in a meaningful sense, beside the point. The selection is the product.
Professional gatekeeping is the function that gives the system its coercive power. You cannot practice medicine, law, engineering, or architecture without passing through university-controlled pipelines. This is defensible in fields where incompetence kills people. It is considerably less defensible when extended to fields where the gatekeeping serves primarily to restrict labor supply and inflate the wages of incumbents. The question of which professions genuinely require years of university training and which have simply captured the credentialing apparatus for guild-like purposes is one that the system has no incentive to ask honestly.
Research is the function that justifies the system’s most extravagant claims about its social value. Universities are, in fact, responsible for an enormous share of basic scientific research, and the tenure system — for all its dysfunction — does provide a structure within which long-term, high-risk intellectual work can occur. But the research function is concentrated in a relatively small number of institutions and involves a relatively small fraction of the faculty. The vast majority of university instructors at the vast majority of institutions are not producing groundbreaking research. They are teaching introductory courses to undergraduates who are there for the credential, not the knowledge.
Cultural reproduction is the function that both progressives and conservatives recognize but describe in incompatible terms. The university transmits a set of assumptions about what constitutes legitimate knowledge, appropriate discourse, and correct values. Whether this is described as “liberal indoctrination” or “critical thinking” depends entirely on whether the observer shares the assumptions being transmitted. The structural point is the same either way: the university is not a neutral platform for the free exchange of ideas. It is an institution with its own orthodoxies, its own taboos, and its own mechanisms for enforcing conformity — mechanisms that are no less powerful for being social rather than formal.
Networking is the function that reveals the entire game. When universities explicitly market “the network” as a core benefit — when alumni connections, career fairs, and “who you’ll meet” feature prominently in admissions brochures — they are saying the quiet part out loud. They are acknowledging that access to social capital, not the transmission of knowledge, is the real product. A student at a state school and a student at an Ivy League institution may take courses of comparable academic rigor. They will not have comparable access to the social networks that determine who gets the interview, who gets the introduction, who gets the funding. The university is selling proximity to power and calling it education. The honesty of the marketing is almost admirable. The implications are damning.
The Economic Model: Two Universities in One
The financial structure of American higher education has produced what is effectively a two-tier system operating under a single institutional label. For students from wealthy families, university is a four-year investment in social capital, personal development, and credential acquisition, funded from existing resources and carrying no financial risk. For students from middle- and lower-income families, the same experience is debt-financed — a leveraged bet on future earnings that may or may not materialize.
These two populations sit in the same lecture halls, but they are not having the same experience. The wealthy student can afford to major in philosophy, take an unpaid internship in a field they find meaningful, spend a semester abroad, and graduate with the freedom to pursue low-paying but high-status work in the arts, nonprofits, or public service. The debt-financed student faces relentless pressure to choose a “practical” major, to work part-time during the semester, to skip the unpaid internship in favor of paid work, and to accept the first salaried position available after graduation regardless of fit or interest. The system offers both students the same diploma. It does not offer them the same freedom to use it.
The aggregate numbers are staggering. Total outstanding student loan debt in the United States exceeds $1.7 trillion. The median borrower takes over twenty years to fully repay. The default rate is highest among students who borrowed relatively small amounts to attend institutions with the lowest completion rates — precisely the population the system claims to be helping. The students who benefit most from the current arrangement are those who needed it least: the already-wealthy, for whom the university is a finishing school, and the exceptionally talented, for whom it is a launching pad. For the broad middle, it is increasingly a financial trap dressed in the language of opportunity.
The pricing model itself has detached from any rational relationship to the service provided. Tuition has increased at roughly three times the rate of inflation for four decades. This is not because the quality of instruction has tripled. It is because the availability of federal student loans has allowed institutions to capture subsidy dollars through tuition increases — a dynamic economists call the Bennett Hypothesis, and one that operates with the same logic as any other market in which a third-party payer insulates the consumer from the true cost. The university raises tuition because it can. The student borrows because they must. The federal government guarantees the loan because the alternative — acknowledging that the system is extractive — is politically unthinkable.
The Institutional Mismatch
The modern university was designed for a world that no longer exists. Its core assumptions — that knowledge is scarce and must be physically co-located with experts, that a career is a stable trajectory requiring front-loaded training, that a four-year residential experience between the ages of eighteen and twenty-two is the optimal structure for intellectual development — were reasonable in the mid-20th century. They are increasingly disconnected from 21st-century reality.
Knowledge is no longer scarce. The lecture that once required physical presence in a specific room at a specific time is now available, often for free, from the world’s leading experts on any internet-connected device. MIT’s OpenCourseWare, Khan Academy, and dozens of similar platforms have demonstrated that the informational content of a university education can be distributed at near-zero marginal cost. What cannot be distributed is the credential, the social network, and the sorting function — which is precisely why universities have not been disrupted by the internet in the way that newspapers, record labels, and travel agencies were. The scarcity is artificial. It is maintained not by the difficulty of transmitting knowledge but by the institutional monopoly on certifying that knowledge has been received.
Careers are no longer stable trajectories. The assumption that a person will train for a profession at twenty, enter it at twenty-four, and practice it until retirement at sixty-five was already fraying in the 1990s. It is now plainly obsolete. The average worker changes jobs every four years. Entire industries emerge and disappear within a single career span. The skills that are valuable today may be irrelevant in a decade. A system designed to front-load all training into a single four-year block at the beginning of adult life is structurally incapable of serving a population that needs continuous, modular, just-in-time learning throughout a forty-year career. The university’s response to this mismatch has been to create continuing education programs and professional master’s degrees — bolting new revenue streams onto the existing structure rather than rethinking the structure itself.
The four-year residential model, meanwhile, persists largely because it serves the social sorting and networking functions that have nothing to do with education. The dormitory, the dining hall, the fraternity, the study-abroad program — these are mechanisms for forming social bonds among a curated peer group during a period of maximum social plasticity. They are extraordinarily effective at what they do. What they do is reproduce class structure with a patina of meritocratic legitimacy. The student who graduates with a degree and a network of well-connected peers has received the actual product. The student who graduates with a degree and $80,000 in debt but no comparable network has received the packaging.
The result is an institution that is simultaneously indispensable and indefensible. Indispensable because the credentialing monopoly means that opting out carries severe economic penalties. Indefensible because the gap between what the institution promises — transformation through knowledge — and what it delivers — sorting by class origin with a four-year delay and a six-figure price tag — grows wider with each tuition increase. The university does not need to be abolished. It needs to be understood for what it is: not a temple of learning that incidentally costs money, but an economic and social sorting mechanism that incidentally involves some learning.
3. The Measurement Failure and the Performance Economy
Why Merit Is Illegible at Scale
The standard critique of meritocracy is that it doesn’t exist — that the game is rigged, the deck is stacked, and the winners are predetermined by birth. This critique contains a great deal of truth, but it misses something structural. Even if every participant started from an identical position, meritocracy would still fail at scale, because merit itself is extraordinarily difficult to measure.
Consider what “real work” actually looks like in most domains. It is slow. It is solitary or conducted in small groups. Its outputs are ambiguous, context-dependent, and often invisible to anyone not deeply embedded in the relevant field. A researcher spending three years pursuing a line of inquiry that ultimately produces a single paper — or no paper at all, because the hypothesis was wrong in an instructive way — is doing real intellectual work. A nurse who prevents a crisis through quiet vigilance and early intervention has contributed enormously, but there is no record of the crisis that didn’t happen. A software engineer who spends six months simplifying a codebase, reducing its complexity by half, has created massive value — but the diff looks like deletion, and deletion does not demo well.
Now consider what performance looks like. It is fast. It is visible. It is legible to people who have no expertise in the domain being performed. A confident presentation, a polished slide deck, a well-timed comment in a meeting, a credential from a recognizable institution, a publication count, a follower count — these are signals that can be read by anyone, evaluated quickly, and compared across individuals without requiring the evaluator to understand the underlying work. Performance is optimized for the observer. Real work is optimized for the problem.
In small groups — a research lab of eight people, a startup of twelve, a workshop of five — the distinction barely matters. Everyone can see what everyone else is doing. The person who talks a good game but produces nothing is identified within weeks. The person who says little but solves the hard problems is recognized and valued. Merit is legible because the evaluators have direct access to the work.
Scale destroys this. In an organization of ten thousand people, or a profession of two million, or a society of three hundred million, no evaluator has direct access to more than a tiny fraction of the work being done. Decisions about hiring, promotion, funding, and recognition must be made by people who cannot possibly observe the relevant work firsthand. They need proxies. And the proxies that are available — credentials, publication metrics, presentation skills, institutional affiliation, visibility, confidence — are precisely the ones that reward performance over contribution.
This is not a conspiracy. It is a measurement problem. Large anonymous systems face a genuine epistemological challenge: how do you identify merit when you cannot observe work? The answer, in practice, is that you substitute legibility for quality. You hire from name-brand schools because evaluating ten thousand applicants on the merits of their actual thinking is impossible. You promote the person who presents well because you cannot directly observe the quality of their daily work. You fund the researcher with the impressive publication count because reading and evaluating the actual content of five hundred papers is beyond any committee’s capacity. Each of these substitutions is individually rational. Collectively, they produce a system that systematically rewards the performance of competence over the possession of it.
The Insecure Performer
This measurement failure produces a specific human type that is ubiquitous in institutional life and almost always misunderstood: the insecure performer. This is the person who has learned — correctly — that survival within large institutions depends more on the appearance of competence than on competence itself. They optimize for visibility. They speak with confidence on subjects they understand superficially. They cultivate relationships with decision-makers. They attach themselves to successful projects at the moment of success and distance themselves from failing ones before the failure becomes visible. They are exquisitely attuned to what the institution rewards and ruthlessly efficient at producing it.
The conventional reading of this person is moral: they are a fraud, a climber, a bullshit artist. This reading is satisfying and almost entirely wrong. The insecure performer is not a character type. They are a product — the predictable output of a system that cannot measure what it claims to value. When the institution says it rewards excellence but actually rewards visibility, the rational actor optimizes for visibility. When the institution says it values deep expertise but promotes generalists who present well, the rational actor becomes a generalist who presents well. When the institution says it cares about outcomes but measures outputs, the rational actor maximizes outputs regardless of outcomes.
The insecurity is the tell. The genuine fraud — the person who knows they are faking and doesn’t care — is relatively rare and relatively easy to identify. Far more common is the person who suspects they are faking, who lives in chronic anxiety about being exposed, who works frantically to maintain the performance precisely because they sense the gap between what they project and what they know. Impostor syndrome is not a psychological disorder. It is an accurate perception of institutional reality: the system selected you for your performance, not your substance, and on some level you know it.
The tragedy is that many of these people are genuinely talented. They entered the system with real ability and real curiosity. But the system trained the curiosity out of them and trained the performance in. After ten or fifteen years of optimizing for legibility — of writing papers designed to be cited rather than to be true, of giving presentations designed to impress rather than to inform, of pursuing projects designed to be visible rather than to be important — the performance has become the person. The mask has eaten the face. They could not return to doing real work even if the incentives changed, because they have forgotten what real work feels like.
This is not a moral failure. It is an institutional one. Blaming individuals for responding rationally to the incentive structures they inhabit is like blaming water for flowing downhill. The question is not “why are there so many performers?” The question is “why did we build a system that selects for performance?”
The Existential Bind
The measurement failure would be merely frustrating if participation in the system were optional. It is not. For most people in developed economies, institutional compliance is not a lifestyle choice — it is a survival requirement. Income, housing, healthcare, professional identity, social standing, and access to the basic infrastructure of modern life are all mediated by institutions that operate according to the logic described above. The person who opts out of the performance economy does not simply forgo status. They risk poverty, homelessness, and the loss of medical care.
This creates what might be called the existential bind: the system demands performance, the individual knows the performance is hollow, but the consequences of refusing to perform are catastrophic. The choice is not between authenticity and inauthenticity. It is between inauthenticity and destitution. When your rent depends on your employer’s assessment of your value, and your employer’s assessment is based on proxies that reward performance over contribution, you perform. When your children’s access to healthcare depends on your continued employment, and your continued employment depends on your visibility within the organization, you optimize for visibility. The bind is not abstract. It is material, immediate, and inescapable for anyone who does not already possess independent wealth.
The phrase that captures this — borrowed from contexts far more extreme but structurally analogous — is “die or conquer the world.” The system presents itself as offering infinite possibility: work hard, develop your talents, and you can achieve anything. The reality is a much narrower corridor: perform according to institutional expectations or face consequences that range from career stagnation to economic ruin. The rhetoric is freedom. The structure is compliance. And the gap between the two produces a specific kind of psychological damage — a pervasive sense that something is wrong, that the game is rigged, that one’s daily activity is disconnected from anything genuinely meaningful — that is then pathologized as individual dysfunction rather than recognized as a rational response to an irrational system.
This is why the discourse around “quiet quitting,” “burnout,” and “disengagement” consistently misses the point. These are not failures of individual motivation. They are symptoms of a measurement failure that has metastasized into an entire economy of performance. The worker who does the minimum is not lazy. They have correctly identified that additional effort will be evaluated not on its merits but on its visibility, and they have decided — consciously or not — that the performance is not worth the energy. The burned-out professional is not weak. They have been running on the fumes of a performance that was never connected to anything they actually cared about, and the fuel has run out. The disengaged employee is not ungrateful. They are experiencing the predictable psychological consequences of spending years optimizing for metrics that have no relationship to the work they were trained to value.
The Measurement Problem Is the Root Problem
The reason this analysis matters — the reason it is not merely an exercise in institutional sociology — is that the measurement failure is upstream of nearly every other dysfunction in the education-to-employment pipeline. Universities reward performance because they cannot measure learning. Employers reward credentials because they cannot measure competence. Promotion committees reward visibility because they cannot measure contribution. Grant agencies reward publication counts because they cannot measure intellectual significance. At every stage, the system substitutes a legible proxy for an illegible reality, and at every stage, the substitution degrades the thing it claims to measure.
This is not a problem that can be solved by better intentions. The administrators, hiring managers, and committee members who rely on proxies are not villains. They are people facing a genuine epistemological problem — how to evaluate quality at scale — and reaching for the only tools available. The tools are inadequate. But the inadequacy is structural, not moral. No amount of diversity training, rubric redesign, or “holistic review” will solve a problem rooted in the fundamental illegibility of merit to large anonymous systems.
The implication is uncomfortable but important: the meritocracy debate is largely beside the point. The question is not whether we have a “true” meritocracy or a “fake” one. The question is whether meritocracy is possible at the scale and complexity of modern institutions. The evidence suggests it is not — not because people are corrupt, but because merit is the kind of thing that can only be recognized by people with direct access to the work, and modern institutions are specifically designed to make decisions without direct access to the work. The system is not failing to be meritocratic. It is succeeding at being something else entirely — a performance economy that sorts people by their ability to appear competent to distant observers — and calling the result meritocracy.
Understanding this is essential context for what follows. The arrival of artificial intelligence does not merely threaten to automate certain tasks. It threatens to disrupt the entire economy of performance by making the proxies that institutions rely on — credentials, confidence, visibility, fluency — trivially easy to produce. If the system was already struggling to distinguish performance from contribution, what happens when the cost of performance drops to zero?
4. The AI Inflection Point: Collapsing Knowledge and Performance
From Expensive to Free to Worthless
The internet made knowledge cheap. AI makes knowledge worthless. These sound like points on the same continuum. They are not. They represent a qualitative shift — a phase transition — in the relationship between knowledge, competence, and economic value.
When the internet democratized access to information in the 1990s and 2000s, it disrupted the distribution of knowledge. The lecture that once required a seat in a specific auditorium could now be watched from anywhere. The textbook that once cost $150 could be found as a PDF. The research paper locked behind a journal paywall could be accessed through Sci-Hub. This was genuinely revolutionary, but it left the underlying value proposition of education largely intact. Knowing something still mattered. Having read the paper still differentiated you from the person who hadn’t. The internet made the raw material of knowledge available to everyone; it did not make everyone capable of using it. The gap between access and mastery remained wide, and institutions could still position themselves as the bridge.
AI closes that gap — not by making people more knowledgeable, but by making knowledge itself non-differentiating. When anyone with access to a large language model can produce a competent legal memo, a structured policy analysis, a literature review with proper citations, a persuasive grant proposal, or a well-reasoned essay on virtually any topic, the possession of the knowledge required to produce those artifacts no longer distinguishes the person who has it from the person who doesn’t. The output is the same. The signal is destroyed.
This is not a hypothetical. It is the current state of affairs. A first-year law student using GPT-4 can produce a brief that is, in many cases, structurally indistinguishable from one produced by a third-year associate who spent three years learning to do it manually. A undergraduate with no background in molecular biology can generate a passable research summary that reads like the work of someone with a graduate education. A person who has never studied economics can produce a policy memo that hits every expected note — framework, evidence, counterargument, recommendation — because the model has internalized the genre conventions of ten thousand such memos. The knowledge hasn’t been transferred. The performance of knowledge has been automated.
This distinction is critical. The internet era threatened institutions that sold access to information — encyclopedias, reference libraries, basic instructional content. Those institutions either adapted or died, but the broader educational edifice survived because it could credibly claim to offer something beyond information: the ability to think with information, to synthesize, to evaluate, to produce. AI threatens institutions that sell the demonstration of that ability. And that, as the previous sections have established, is what most educational institutions actually sell.
The Simulation of Competence
What makes AI qualitatively different from previous technological disruptions is its ability to simulate the surface markers of competence — precisely the markers that institutions have spent decades using as proxies for the real thing.
Consider what a university education is supposed to produce: a person who can write clearly, argue logically, synthesize information from multiple sources, adopt the appropriate register for a given audience, and present ideas in a structured, professional format. Now consider what a large language model does: it writes clearly, argues logically, synthesizes information from multiple sources, adopts the appropriate register for a given audience, and presents ideas in a structured, professional format. The overlap is not coincidental. The model was trained on the outputs of educated humans. It has learned, with remarkable fidelity, to reproduce the performance layer of education — the visible, assessable, legible markers that institutions use to certify that education has occurred.
This is devastating not because AI is smarter than educated humans — in many important respects, it is not — but because the system was never actually measuring intelligence or understanding. It was measuring performance. And performance is exactly what AI is optimized to produce. The polished essay, the well-structured argument, the professional tone, the appropriate use of evidence, the confident synthesis of complex material — these were the signals that separated the educated from the uneducated, the credentialed from the uncredentialed, the competent from the incompetent. When a machine can produce those signals on demand, at zero marginal cost, for anyone who asks, the signals cease to function. They become noise.
The implications cascade through every institution that relies on the performance layer. The university that assesses students through essays and exams is assessing artifacts that AI can produce. The employer that screens candidates through writing samples and case analyses is screening for a capability that AI provides to everyone equally. The professional who justifies their salary through the production of reports, analyses, and recommendations is justifying it through outputs that a model can generate in seconds. The entire apparatus of knowledge-based credentialing — the mechanism by which society sorts people into economic roles based on demonstrated intellectual capability — is undermined not by making people smarter but by making the demonstration of intelligence trivially reproducible.
The Collapse of the Performance Layer
Recall the insecure performer from the previous section: the person who has learned to optimize for the appearance of competence because the system cannot measure competence directly. AI does not merely compete with this person. It annihilates their entire value proposition.
The insecure performer’s skill set — the ability to produce polished outputs, to speak fluently about topics understood only superficially, to assemble impressive-looking deliverables from templates and conventions, to project confidence and authority through the mastery of professional genre — is precisely the skill set that AI replicates most effectively. The model is, in a sense, the ultimate insecure performer: it produces flawless surfaces with no underlying understanding. It can write a McKinsey-style strategy deck without knowing what a business is. It can draft a legal argument without understanding justice. It can produce a research proposal without caring whether the research matters. It is pure performance, distilled and automated.
For the person whose career was built on performance rather than contribution, this is an extinction-level event. Not because AI will take their job tomorrow — organizational inertia is slower than that — but because the thing they spent years learning to do is now available to everyone for free. The junior analyst who spent two years learning to format reports in the house style. The mid-level manager who built a reputation on well-crafted emails and articulate meeting summaries. The consultant whose value lay in the ability to produce authoritative-sounding documents on topics they’d researched for seventy-two hours. The academic who published prolifically by mastering the conventions of their field’s journals without ever producing a genuinely original idea. Each of these people built their career on a form of performance that AI can now replicate at scale, instantly, and for nothing.
The people who are not threatened — or who are less threatened — are those whose value was never in the performance layer to begin with. The researcher who actually understands the system they study well enough to know when the model is wrong. The engineer who can look at an AI-generated design and identify the subtle physical constraint it has ignored. The writer whose value lies not in grammatical competence but in having something to say that no model would generate because no model has lived a human life. The teacher who can recognize the difference between a student who understands and a student who has pasted. The doctor whose diagnostic skill comes from twenty years of pattern recognition in the presence of ambiguity that no training dataset captures. These people have something that cannot be simulated because it was never a performance in the first place. It was — and remains — the real thing.
But here is the structural problem: the system cannot tell the difference. The same measurement failure that allowed insecure performers to thrive now prevents institutions from distinguishing between AI-augmented performance and genuine human competence. If you couldn’t tell the difference between a person who understood the material and a person who performed understanding before AI, you certainly cannot tell the difference now that the performance has been perfected by a machine. The evaluative infrastructure — grades, credentials, writing samples, interviews, presentations — was already inadequate. AI has not created the measurement problem. It has made the measurement problem unsolvable within the existing framework.
Why This Is Different
Every generation of educators has faced a technological disruption and declared it existential. The calculator would destroy mathematical reasoning. The internet would make students unable to remember facts. Wikipedia would eliminate the need for research skills. In each case, the institution adapted — sometimes grudgingly, sometimes clumsily — and survived. The reasonable person might conclude that AI is simply the latest in this series: alarming at first, ultimately absorbed.
This conclusion is wrong, and understanding why it is wrong requires understanding what was actually at stake in each previous disruption.
The calculator automated computation — a mechanical process that was never the core value proposition of mathematical education. Teachers could (and did) shift emphasis to conceptual understanding, problem formulation, and mathematical reasoning. The thing the calculator replaced was not the thing the institution was selling.
The internet automated access to information — but left intact the skills required to evaluate, synthesize, and deploy that information. Teachers could (and did) shift emphasis from memorization to critical analysis. The thing the internet replaced was not the thing the institution was selling.
AI automates the demonstration of understanding itself. It produces the essays, the analyses, the arguments, the syntheses — the very outputs that institutions use to certify that understanding has occurred. There is no further retreat available. When the calculator arrived, educators could say “the calculation isn’t the point; the reasoning is the point.” When the internet arrived, educators could say “the information isn’t the point; the analysis is the point.” When AI arrives and produces the reasoning and the analysis, what is left to retreat to?
The honest answer is: the things that were always actually valuable but that the system was never designed to measure. Genuine understanding. Original insight. The ability to know when a plausible-sounding answer is wrong. The capacity to formulate questions that no one has thought to ask. Judgment refined by experience. Wisdom accumulated through failure. These are real, and they are human, and they are — for now — beyond AI’s reach. But they are also precisely the things that the educational system has never been able to assess at scale, which is why it relied on the performance proxies that AI has just rendered worthless.
The institution is not facing a challenge to its methods. It is facing a challenge to its epistemic foundation — the assumption that the ability to produce certain outputs is evidence of certain internal states. That assumption was always questionable. AI has made it untenable.
The Institutional Response Problem
The predictable institutional response to AI is already visible: ban it, detect it, regulate it, pretend it doesn’t change anything fundamental. Universities are investing in AI detection software. Employers are adding “no AI” clauses to assignments. Professional organizations are drafting policies on “acceptable use.” These responses are understandable and futile. They are the equivalent of the music industry suing Napster users — a rearguard action by institutions that have correctly identified the threat but incorrectly believe it can be contained by enforcement.
AI detection is an arms race that the detectors will lose. Each generation of models produces output that is more human-like, more varied, more resistant to statistical detection. The student who uses AI clumsily — pasting raw output without editing — will be caught. The student who uses AI skillfully — generating ideas, restructuring arguments, polishing prose, integrating AI-assisted passages with their own writing — will be indistinguishable from the student who did it all manually. And the distinction itself is becoming incoherent. When a student uses AI to brainstorm, then writes their own draft, then uses AI to identify weaknesses, then revises — at what point did the work stop being “theirs”? The question has no principled answer because the boundary between tool use and fraud was always a social convention, not a natural kind, and AI has made the convention unenforceable.
The deeper problem is that the institutional response assumes the goal is to preserve the existing assessment framework — to find new ways of verifying that students can do what AI can do, without AI’s help. But this is precisely backwards. The question is not “how do we prevent students from using AI to produce essays?” The question is “why are we assessing students by asking them to produce essays that AI can produce better and faster?” If the output can be generated by a machine, then the ability to generate it manually is no longer a meaningful measure of anything except the willingness to do unnecessary work. The institution that responds to AI by doubling down on AI-proof assessment is an institution that has confused its methods with its mission.
What would a genuine response look like? It would start with an honest reckoning: the things we were measuring were never the things we claimed to value. We said we valued understanding, but we measured output. We said we valued critical thinking, but we assessed essay structure. We said we valued intellectual growth, but we counted credits. AI has not broken the system. It has revealed that the system was already broken — that the relationship between what we measure and what we care about was always tenuous, and that we were relying on the difficulty of producing good output as a rough-and-ready proxy for the presence of genuine understanding. Now that the difficulty has been removed, the proxy has collapsed, and we are left with the question we should have been asking all along: what does it actually mean to be educated, and how would we know if someone were?
That question is not rhetorical. It has answers — or at least, it has better and worse approaches. But the answers require abandoning the assessment frameworks that institutions have spent decades building, the credentialing structures that entire industries depend on, and the comfortable fiction that a degree certifies knowledge rather than compliance. The institutional will to undertake that abandonment does not currently exist. The institutional necessity to undertake it is approaching rapidly.
5. The Disaggregation of Universities
The Bundling Logic and Its Expiration
The pattern is familiar from every other industry the internet has restructured. Newspapers were bundles — classifieds, sports scores, weather, opinion columns, investigative journalism, and comic strips, bound together by the economics of print distribution. The bundle made sense when the cost of printing and delivering a physical object meant that each component subsidized the others. Craigslist took the classifieds. ESPN took the scores. Weather.com took the forecast. The opinion columns migrated to blogs, then to Substack. What remained was the most expensive and least commercially viable component — investigative journalism — stripped of the revenue streams that had quietly funded it for a century. The newspaper didn’t die because people stopped wanting news. It died because the bundle came apart, and the pieces were worth more separately than together — to everyone except the institution that had depended on their combination.
Amazon did the same thing to retail. Wikipedia did it to encyclopedias. Spotify did it to the album. In each case, a legacy institution had bundled together functions that were logically distinct but economically interdependent. In each case, technology made it possible to unbundle those functions and serve them individually, more cheaply, more conveniently, and at greater scale. In each case, the institution that had maintained the bundle insisted that the components were inseparable — that you couldn’t have the classifieds without the journalism, the hit single without the deep cuts, the encyclopedia entry without the editorial apparatus. In each case, the market disagreed.
The university is the next bundle to come apart. Not because anyone has decided to dismantle it, but because the same structural logic applies. The university bundles at least six distinct functions — credentialing, networking, professional socialization, research, cultural reproduction, and the coming-of-age transition — into a single four-year residential package. This bundle made sense when knowledge was scarce, when social networks could only be formed through physical co-presence, when credentials required institutional authority to be legible, and when the transition from adolescence to adulthood needed a structured container. Each of those conditions is eroding. The bundle is losing its binding logic.
Universities will not disappear. Newspapers still exist. Retail stores still exist. Encyclopedias still exist. But they exist as diminished versions of what they were, serving narrower functions for smaller audiences at reduced scale. The institution persists; the monopoly does not. What is happening to universities is not destruction. It is decomposition — the slow unbundling of functions that were never logically unified, each migrating toward whatever structure serves it most efficiently.
The Credential Function → Reputation Networks and Verifiable Contribution Graphs
The credential was always the weakest link in the bundle, because it was always the most dishonest. A bachelor’s degree does not certify that you know anything specific. It certifies that you were admitted to an institution (sorting), that you remained enrolled for four years (compliance), and that you accumulated enough credits to graduate (persistence). The knowledge component is, at best, loosely correlated with the credential. Everyone in the system knows this. Employers know it. Graduates know it. The fiction is maintained because no better alternative existed — because the cost of evaluating each individual on their actual merits was prohibitive, and the degree served as a cheap, if noisy, signal.
AI collapses the cost of that evaluation. Not immediately, and not completely, but directionally and irreversibly. The same technology that makes the performance of competence trivially reproducible also makes the verification of actual competence newly possible. Verifiable contribution graphs — persistent, public records of what a person has actually built, written, solved, designed, or shipped — are already emerging as alternatives to the credential. Open-source contribution histories on GitHub. Portfolios of shipped products. Public writing with traceable intellectual development over time. Peer attestations within professional communities. On-chain records of project participation and outcome.
None of these are new in concept. What is new is the infrastructure to make them legible at scale. When an employer can review a candidate’s actual work product — not a résumé’s description of work product, but the work itself, with provenance, context, and peer evaluation attached — the degree becomes what it always secretly was: a proxy for class background and institutional access, not a measure of capability. The proxy doesn’t vanish overnight. Proxies are sticky. But it loses its monopoly grip on the first-pass filter, and that is enough to begin the unbundling.
The institutions that will thrive in this transition are those that help people build legible track records of real contribution rather than those that stamp diplomas. The credential doesn’t die. It decomposes into something more granular, more continuous, and more honest: not “this person attended our institution for four years” but “here is what this person has actually done, verified by the people who worked alongside them.”
The Networking Function → Guilds and High-Trust Micro-Communities
The networking function is the one universities are most honest about and least replaceable in. When an admissions brochure says “join a community of extraordinary peers,” it is making a promise that has nothing to do with education and everything to do with social capital. And it is a promise the institution can keep, because the admissions process itself — the sorting — creates the curated peer group that makes the network valuable. You are not paying for the lectures. You are paying for the other people in the room.
This function is genuinely valuable. The question is whether it requires a four-year, six-figure residential program to deliver. Increasingly, the answer is no — not because the function has become less important, but because the mechanisms for forming high-trust professional communities have multiplied.
Guilds — in the medieval sense of professional communities organized around shared craft, mutual accountability, and graduated levels of mastery — are re-emerging in digital form. Small, curated communities organized around specific domains: AI safety researchers who know each other’s work and vouch for each other’s competence. Independent software developers who collaborate on shared infrastructure and build reputations through visible contribution. Writers, designers, and strategists who form collectives that function as both professional networks and quality-assurance mechanisms. These communities are smaller than university alumni networks, but they are denser — the connections are based on shared work rather than shared geography, and the trust is earned through demonstrated competence rather than conferred by institutional affiliation.
The key structural difference is that these micro-communities are organized around function rather than age cohort. A university network is formed during a four-year window when participants are eighteen to twenty-two, and it is maintained (or not) for decades afterward. A guild is formed around ongoing shared practice and is continuously refreshed by new participants who demonstrate relevant capability. The university network’s value depreciates as members’ careers diverge. The guild’s value appreciates as members’ expertise deepens. The university gives you a network once. The guild gives you a network that evolves with your work.
This does not mean university networks become worthless. The Harvard alumni network will retain its value for decades, because its value was never about competence — it was about access to power, and power is self-reinforcing. But for the vast majority of professionals who attended institutions without world-historic brand recognition, the alumni network was always a weak substitute for the genuine professional community they actually needed. Those people are increasingly finding that community elsewhere.
The Professional Socialization Function → Apprenticeship Pipelines
One of the university’s least discussed but most important functions is professional socialization — the process by which a person learns not just the knowledge of a field but its culture: how practitioners think, what they value, how they communicate, what counts as good work, what the unwritten rules are. Medical school does not merely teach anatomy. It teaches students how to be doctors — how to manage uncertainty, how to communicate with patients, how to navigate the hierarchy of a hospital, how to make decisions under conditions of incomplete information. Law school does not merely teach legal doctrine. It teaches students how to think like lawyers — how to read a case, how to construct an argument, how to identify the issue that matters in a sea of irrelevant facts.
This socialization function is real and important. It is also wildly inefficient as currently delivered. Three years of law school to learn professional socialization that could be transmitted in six months of structured apprenticeship. Four years of undergraduate education followed by two years of business school to learn management practices that are better learned by managing. The university stretches the socialization process to fill the time required by the credential, padding it with coursework that serves the institution’s revenue model rather than the student’s professional development.
Apprenticeship pipelines — structured programs that embed learners directly in professional practice under the guidance of experienced practitioners — deliver the socialization function more efficiently, more honestly, and with better outcomes. This is not a theoretical claim. It is the observed reality in every field where apprenticeship models exist alongside university models. Software engineers trained through bootcamps and on-the-job mentorship reach professional competence faster than those who complete four-year computer science degrees. Tradespeople trained through apprenticeship programs outperform those with classroom-only preparation. The medical residency — which is itself an apprenticeship — is universally acknowledged to be where doctors actually learn to practice medicine, not the preceding years of classroom instruction.
The barrier to apprenticeship at scale has always been the same: it requires experienced practitioners to invest time in training novices, and that investment is costly. But AI changes the economics. When a large language model can handle the routine knowledge-transfer component — answering factual questions, explaining concepts, providing examples, offering feedback on basic work product — the human mentor is freed to focus on the things that actually require human judgment: modeling professional reasoning, providing context-sensitive feedback, transmitting tacit knowledge, and evaluating the learner’s developing judgment in situations where the right answer is ambiguous. AI doesn’t replace the apprenticeship. It makes the apprenticeship scalable by automating the parts that never needed a human in the first place.
The Research Function → Hybrid Human-AI Cognitive Institutions
The research function is the one that most clearly justifies the university’s existence and the one that is most poorly served by the university’s current structure. The tenure system was designed to protect intellectual freedom — to give researchers the security to pursue long-term, high-risk work without fear of institutional reprisal. In practice, it has produced a gerontocracy in which senior researchers control resources, junior researchers compete for scraps, and the incentive structure rewards prolific publication of incremental results over the slow, risky pursuit of genuine breakthroughs.
AI does not solve this problem, but it does create the conditions for new institutional forms that might. The emerging model is what might be called the hybrid cognitive institution — a research organization designed from the ground up around the collaboration between human researchers and AI systems. These institutions look nothing like traditional university departments. They are smaller, flatter, more interdisciplinary, and organized around problems rather than disciplines. They use AI not as a tool for automating existing research workflows but as a cognitive partner that extends the researcher’s capacity for pattern recognition, hypothesis generation, literature synthesis, and experimental design.
The key structural difference is that these institutions can be built without the overhead that makes universities so expensive and so slow. A hybrid research lab does not need a campus, a football team, a residential housing system, a dining hall, or an administrative apparatus employing more people than the research faculty. It needs researchers, computing infrastructure, and a governance structure that protects intellectual freedom without requiring the baroque tenure process that universities have evolved. The cost structure is radically different, which means the funding model can be radically different — less dependent on federal grants administered through bureaucratic review processes, more amenable to philanthropic funding, industry partnerships, and novel mechanisms like retroactive public goods funding.
This does not mean university research disappears. The large-scale infrastructure that only universities and national labs can maintain — particle accelerators, telescope arrays, longitudinal population studies — will continue to require institutional scale. But the median research project — the kind of work that constitutes the bulk of academic output — does not require that scale. It requires smart people, good tools, and the freedom to follow interesting questions. Those conditions can be created outside the university, and increasingly, they are.
The Coming-of-Age Function → Intentional Communities and Cohort-Based Programs
The function that is hardest to discuss honestly is the one that has the least to do with education: the university as a coming-of-age ritual. For millions of eighteen-year-olds, college is the first experience of living away from their parents, managing their own time, forming their own identity, navigating sexual and social relationships without parental oversight, and discovering — through trial and error, often painful — who they are and what they care about. This is not a trivial function. It is, for many people, the most important thing that happens during their college years. And it has essentially nothing to do with the curriculum.
The university stumbled into this function by accident. The residential model was designed for pedagogical reasons — to immerse students in a learning environment — but its most powerful effects are social and developmental. The dormitory, the late-night conversation, the student organization, the romantic entanglement, the first encounter with people from radically different backgrounds, the freedom to fail without catastrophic consequences — these experiences constitute a structured transition from adolescence to adulthood that many cultures have historically provided through other means (military service, religious pilgrimage, apprenticeship in a distant city) but that modern secular society has largely outsourced to the university.
The problem is that this function is bundled with a credential that costs $200,000 and takes four years to acquire. The coming-of-age experience does not require four years. It does not require a campus. It does not require professors or courses or exams. It requires a curated cohort of peers, a structured environment with enough freedom to allow genuine exploration and enough support to prevent catastrophe, and a defined period of transition with a clear beginning and end.
Intentional communities and cohort-based programs are beginning to provide this function unbundled from the credential. Gap year programs that place young people in structured cohorts for six months to a year. Residential programs organized around specific themes — environmental stewardship, civic engagement, creative practice — that provide the social container without the academic apparatus. Co-living arrangements designed for young adults in transition, with built-in mentorship and community structure. These programs are small, scattered, and inconsistent in quality. They are also growing, because they address a real need that the university addresses only incidentally and at enormous cost.
The cultural resistance to this unbundling is fierce, because the coming-of-age function is wrapped in nostalgia and identity. For the professional class, “going to college” is not merely an educational decision — it is a rite of passage, a marker of social arrival, a shared experience that binds generations. Suggesting that the developmental function could be served by something other than a four-year university is, for many families, not an argument about institutional efficiency but an attack on a deeply held cultural narrative about what it means to grow up, to succeed, to belong. This resistance is real and should not be dismissed. But it should be recognized for what it is: a cultural attachment to a specific form, not an argument for the form’s necessity.
Decomposition, Not Destruction
The university does not disappear. It decomposes. Each function migrates toward the institutional form that serves it most efficiently, most honestly, and at the most appropriate scale. Credentials become continuous and contribution-based. Networks become functional and practice-based. Professional socialization becomes apprenticeship. Research becomes hybrid and distributed. The coming-of-age transition becomes intentional and unbundled from the credential.
What remains of the university after this decomposition is not nothing. It is the residual institution that serves the functions that genuinely require institutional scale and co-location: large-scale research infrastructure, the preservation and transmission of cultural heritage, the maintenance of intellectual communities in fields too small or too uncommercial to sustain themselves in the market. These are real and important functions. They are also much smaller than the current university, much less expensive, and much more honest about what they are.
The analogy to newspapers is instructive in its specifics. The New York Times still exists. It is a real institution doing real journalism. But it is not the newspaper industry of 1995. It does not bundle classifieds with investigative reporting. It does not serve as the default information source for an entire metropolitan area. It has found a sustainable model — digital subscriptions, a global audience, a focus on the journalism that justifies its existence — but that model supports a fraction of the journalists, at a fraction of the revenue, serving a fraction of the functions that the old bundle served. The rest of those functions are served elsewhere, by different institutions, in different forms, often better and cheaper than the newspaper ever managed.
The university of 2045 will look like the New York Times of 2025: still real, still valuable, still prestigious in some cases, but no longer the default container for a bundle of functions that have found better homes elsewhere. The question is not whether this decomposition will happen. The structural pressures are too strong and the alternatives too numerous for the bundle to hold. The question is whether the transition will be managed with some degree of intentionality and care for the people caught in it — the students currently taking on debt for a bundle that is mid-decomposition, the faculty whose careers are built on institutional forms that are losing their logic, the communities whose economies depend on institutions that are slowly hollowing out — or whether it will be left to the market’s indifferent efficiency, which optimizes for cost and convenience and has no mechanism for preserving the things that were genuinely valuable about the institution it is replacing.
6. The New Scarcities and Micro-Meritocracies
What Becomes Scarce When Knowledge Is Free
If the previous sections describe what AI destroys — the performance layer, the credential signal, the bundling logic of the university — this section asks the harder question: what does AI make more valuable?
The answer is not “nothing.” It is a specific set of capacities that share a common feature: they cannot be faked, cannot be automated, and cannot be acquired without sustained investment of time and attention in direct contact with reality. These are the new scarcities, and they will define the next economy of human value.
Judgment — the ability to make good decisions under genuine uncertainty, where the relevant information is incomplete, contradictory, or absent. A language model can synthesize everything that has been written about a problem. It cannot tell you what to do when the problem has no precedent, when the data points in conflicting directions, and when the consequences of being wrong are irreversible. Judgment is not analysis. It is the thing that remains after analysis has been exhausted.
Taste — the cultivated ability to distinguish the significant from the trivial, the excellent from the merely competent, the genuine from the performed. In a world where competent output is infinite and free, the ability to recognize which output matters — which research question is worth pursuing, which design is elegant rather than merely functional, which argument is true rather than merely valid — becomes the binding constraint. Taste is not preference. It is discernment refined through years of exposure to excellence and failure.
The ability to know what matters — a meta-capacity that encompasses judgment and taste but extends beyond them. It is the capacity to look at a complex situation and identify the one thing that will make the difference. The doctor who, amid a cascade of symptoms and test results, sees the pattern that points to the actual diagnosis. The engineer who, amid a thousand possible optimizations, identifies the one constraint that is actually binding. The leader who, amid competing demands and political pressures, recognizes the decision that cannot be deferred. This capacity is not teachable in the conventional sense. It is developed through years of practice, failure, reflection, and mentorship — through the slow accumulation of pattern recognition that no shortcut can replicate.
Relational depth — the ability to build and maintain the kind of trust that makes genuine collaboration possible. AI can coordinate tasks. It cannot build the mutual understanding between two people who have worked together long enough to communicate in shorthand, to anticipate each other’s reasoning, to know when the other person’s hesitation means “I see a problem you don’t.” This kind of relational capital is built slowly, cannot be transferred, and becomes more valuable as the transactional layers of work are automated away.
These scarcities share a structural feature that distinguishes them from the competencies the old system measured: they are illegible at scale. You cannot test for judgment on a standardized exam. You cannot certify taste with a credential. You cannot verify the ability to know what matters through a portfolio review. These capacities are visible only to people who possess them — who can recognize them in others because they know what they look like from the inside. This is why the emerging institutional forms are small: not because small is fashionable, but because the things that matter can only be evaluated by people with direct access to the work.
The Micro-Meritocracy
If merit is illegible at scale but legible in small groups, the structural implication is clear: genuine meritocracy is possible only in communities small enough for mutual observation. This is the micro-meritocracy — a social structure in which evaluation is based on direct knowledge of a person’s work, judgment, and character, rather than on proxies legible to distant observers.
The micro-meritocracy is not a new invention. It is the oldest form of professional organization. The medieval guild, the research laboratory, the surgical team, the jazz ensemble, the startup of eight people in a garage — these are all micro-meritocracies. In each, the members know each other’s capabilities through direct observation. The person who can do the work is recognized. The person who cannot is identified. The evaluation is not perfect — personal biases, social dynamics, and power imbalances all distort it — but it is connected to reality in a way that large-scale proxy-based evaluation never can be.
What is new is the technological infrastructure that allows micro-meritocracies to network — to connect with each other, to share reputational information across community boundaries, and to create legibility at scale without sacrificing the direct observation that makes the evaluation honest. A guild of forty software engineers can evaluate its own members with high fidelity. When that guild’s assessments are visible to other guilds, to employers, and to the broader professional ecosystem — through contribution graphs, peer attestations, and shared project histories — the micro-meritocracy’s evaluations become portable without becoming diluted.
This is the structural alternative to the credential: not a single institution certifying competence from above, but a network of small communities certifying competence from within, with technological infrastructure making those certifications legible across community boundaries. The unit of trust is not the institution. It is the relationship. And the relationships are organized not by geography or age cohort but by shared practice and demonstrated capability.
The limitations are real. Micro-meritocracies can be insular, exclusionary, and prone to groupthink. They can reproduce existing inequalities by admitting only people who already resemble current members. They can become echo chambers in which a narrow definition of excellence crowds out genuine diversity of thought and approach. These are not hypothetical risks — they are the observed failure modes of every small, high-trust community in history. The guild that evaluates its members honestly may also evaluate outsiders unfairly. The research lab that recognizes genuine talent may also be blind to talent that doesn’t match its existing template.
But these failure modes, severe as they are, differ in kind from the failure modes of large-scale proxy-based systems. The micro-meritocracy fails by being too narrow. The macro-system fails by being disconnected from reality entirely. The micro-meritocracy’s biases are at least correctable — visible to members, subject to internal challenge, amenable to reform through the admission of new perspectives. The macro-system’s biases are structural and self-reinforcing — embedded in the proxies themselves, invisible to the people who rely on them, and resistant to reform because the people who benefit from the biases are the ones who would have to identify and correct them.
The transition from macro-credentialing to networked micro-meritocracies is not a clean upgrade. It is a shift from one set of failure modes to another — from a system that is reliably mediocre to a system that is unevenly excellent. Whether this constitutes progress depends on whether the new system’s failure modes can be mitigated more effectively than the old system’s, and on whether the people who are currently served (however poorly) by the old system will be served at all by the new one. These are open questions, and they lead directly to the concerns that follow.
7. Open Questions and Failure Modes
Everything above describes structural pressures that are real, directional, and — in the aggregate — probably irreversible. But describing a trajectory is not the same as endorsing it, and identifying what is likely is not the same as identifying what is good. The transition from the current educational regime to whatever replaces it is not a clean upgrade. It is a period of institutional decomposition during which millions of people will be navigating systems that are half-old and half-new, with the old structures losing their logic but retaining their coercive power, and the new structures gaining coherence but lacking the infrastructure to catch the people who fall between them.
The following are not rhetorical questions. They are genuine gaps in the analysis — places where the structural logic points in a clear direction but the human reality is unresolved, where the theory is clean but the implementation is treacherous, and where the failure modes are severe enough to warrant naming explicitly rather than hoping they resolve themselves.
How Do People Cross the Bridge?
The transition from credential-based sorting to reputation-based sorting sounds liberating in the abstract. In practice, it requires individuals to get from one side to the other without drowning in the middle.
Consider the twenty-two-year-old from a lower-middle-class family who, under the current system, takes on $60,000 in debt to acquire a degree that functions as an entry ticket to the professional economy. The degree is overpriced, the education is often mediocre, and the debt burden is punishing — but the ticket works. Employers accept it. The HR filter passes it. The person gets a job, begins repaying, and enters the economic mainstream. The system is extractive, but it is legible. Everyone understands the rules.
Now imagine telling that person: don’t take on the debt. Instead, build a portfolio of real work. Contribute to open-source projects. Write publicly. Develop a verifiable track record. Join a guild. Earn your reputation through demonstrated competence rather than institutional affiliation.
This advice is structurally sound and practically catastrophic for anyone who does not already possess a safety net. Building a reputation takes time — years, typically. During those years, the person needs to eat, pay rent, and maintain health insurance. The credential-based system, for all its flaws, provides a structured on-ramp: you enroll, you attend, you graduate, you get a job. The reputation-based system has no comparable on-ramp. It assumes a period of investment — of working for visibility rather than income, of building a track record before the track record pays — that is available only to those who can afford to wait.
This is not a minor implementation detail. It is the central equity problem of the entire transition. If the new system’s entry mechanism requires precisely the kind of financial cushion that the old system’s debt financing was designed (however poorly) to provide, then the transition does not democratize opportunity. It re-aristocratizes it. The children of the wealthy build portfolios during gap years funded by their parents. The children of the poor take on debt for a depreciating credential because they cannot afford the luxury of building a reputation on spec.
What would a genuine bridge look like? Stipended apprenticeship programs? Universal basic income during a transition period? Employer-funded reputation-building pipelines that replace tuition reimbursement? Income-share agreements tied to guild membership rather than institutional enrollment? Each of these has been proposed. None has been implemented at scale. The structural incentives for building the bridge are weak, because the people who benefit most from the new system are precisely those who need the bridge least, and the people who need the bridge most have the least political power to demand it.
This is an open question, not a solved problem. And the stakes are not abstract. Every year the transition advances without a viable bridge, another cohort of young people from non-wealthy backgrounds faces a choice between a credential that is losing its value and a reputation economy they cannot afford to enter. The window during which both systems coexist — during which the old credential still works but the new reputation networks are already forming — is the window during which the bridge must be built. If it is not built during that window, the result will not be a more meritocratic system. It will be a more stratified one.
How Do You Verify “Real Work” in a World of Synthetic Output?
The argument for reputation-based sorting depends on a critical assumption: that real work can be distinguished from performed work, and that contribution graphs, portfolios, and peer attestations provide a more honest signal than credentials. This assumption was reasonable before generative AI. It is now under severe pressure.
If a large language model can produce a competent essay indistinguishable from a human-written one, it can also produce a competent open-source contribution, a competent design portfolio, a competent policy memo, a competent research summary. The same technology that destroyed the credential’s signal-to-noise ratio threatens to destroy the portfolio’s signal-to-noise ratio. The same measurement problem reasserts itself one level up: if you couldn’t tell whether the essay reflected genuine understanding, how will you tell whether the GitHub commit history reflects genuine engineering skill?
Cryptographic verification — signing commits, timestamping contributions, recording peer reviews on-chain — addresses provenance but not quality. It can prove that a person submitted a piece of code. It cannot prove that the person understood the code, that the code solved a real problem, or that the person didn’t generate it with a model and submit it with minor edits. The verification problem is not “who produced this artifact?” It is “does this artifact reflect the kind of understanding and judgment that we actually care about?” And that question is no easier to answer for a portfolio than it was for a diploma.
The most promising approaches involve process verification rather than output verification — evaluating how someone works rather than what they produce. Pair programming sessions. Live problem-solving under observation. Iterative projects with documented decision histories that reveal reasoning, not just results. Peer evaluation by people embedded in the same working context who can distinguish between the person who understands the system and the person who queries the model effectively. These approaches work. They also do not scale. They require exactly the kind of direct observation that large anonymous systems cannot provide, which is why the proxy problem existed in the first place.
Is there a verification architecture that is simultaneously resistant to AI-generated noise, scalable beyond small communities, and accessible to people without existing network connections? This is not a rhetorical question. It is a genuine unsolved problem, and the viability of the entire reputation-based alternative depends on its answer. If the answer is no — if verification necessarily requires small-scale, high-trust, direct-observation contexts — then the reputation economy will work beautifully for people already embedded in such contexts and will be functionally inaccessible to everyone else. Which is to say: it will reproduce the exclusivity of the old system in a new form.
How Is Taste Cultivated Outside the Academy?
The analysis argues that the new scarcities — judgment, taste, the ability to know what matters — are the durable human advantages in an AI-saturated world. This is probably correct. But it raises a question that deserves more scrutiny: where do these capacities come from?
Taste — in the sense used here, meaning the cultivated ability to distinguish the significant from the trivial, the genuine from the performed, the excellent from the merely competent — is not innate. It is developed through sustained exposure to excellence, guided by people who already possess it, in environments that reward discernment over production. Historically, the liberal arts university was one of the primary environments in which this cultivation occurred. The seminar room, the close reading of difficult texts, the sustained argument with a professor who would not accept a superficial answer, the four-year immersion in a community that valued intellectual seriousness — these were the conditions under which taste was formed.
If the university is decomposing, what replaces this function? The autodidact’s answer — “read great books on your own, think carefully, develop your own judgment” — is true for a small number of exceptionally self-directed individuals and irrelevant for the vast majority of people, who develop taste through social processes: mentorship, apprenticeship, immersion in communities where standards are high and feedback is honest. The internet provides access to the raw materials of taste — every great book, every great film, every great piece of music is available — but access to materials is not the same as cultivation. A person with access to the Louvre’s entire collection online is not in the same position as a person who spent four years studying art history with a demanding teacher who could explain why this painting matters and that one doesn’t.
The guild model described earlier may partially address this — small communities of practitioners who transmit standards through shared work and honest critique. But guilds are organized around practice, not around the broader cultivation of judgment that the liberal arts tradition aspired to. A guild of software engineers can cultivate taste in code. It is less clear that it can cultivate the kind of cross-domain judgment — the ability to see connections between fields, to draw on history and philosophy and literature when making decisions about technology or policy — that the best liberal arts education provided.
This is not an argument for preserving the university in its current form. It is an observation that the decomposition of the university creates a genuine gap in the cultivation of a capacity that the analysis itself identifies as increasingly important. If taste is the new scarcity, and if taste requires cultivation in specific social and intellectual conditions, then the question of how those conditions are created outside the traditional academy is not peripheral. It is central. And it does not yet have a convincing answer.
How Will Legacy Institutions Fight Back?
The analysis to this point has treated institutional decomposition as a structural inevitability — the result of technological and economic pressures that no institution can resist indefinitely. This is probably correct in the long run. But institutions do not go quietly, and the long run can be very long indeed.
Legacy universities possess enormous reserves of social capital, political influence, and cultural authority. They control professional licensing pipelines. They sit on endowments measured in tens of billions of dollars. Their alumni occupy positions of power in government, finance, media, and technology. They have centuries of accumulated prestige that functions as a self-reinforcing asset: the degree is valuable because employers value it, and employers value it because it is valuable. Breaking this loop requires not just a better alternative but an alternative so overwhelmingly superior that it can overcome the coordination problem — the fact that no individual employer wants to be the first to stop requiring degrees, and no individual student wants to be the first to forgo one.
The predictable institutional responses are already visible. Universities will absorb the language of the new system — “portfolios,” “competency-based assessment,” “real-world experience” — while preserving the structural logic of the old one. They will offer “micro-credentials” that are really just smaller, more frequent versions of the same credential, still requiring enrollment, still requiring tuition, still controlled by the institution. They will partner with employers to create “apprenticeship programs” that are really just internships with a new label, still gated by admissions, still bundled with coursework, still generating revenue for the university. They will adopt AI tools while lobbying for regulations that make AI-based alternatives to university education harder to operate. They will invoke quality control, consumer protection, and academic standards — all legitimate concerns — as justifications for regulatory frameworks that happen to protect their market position.
More subtly, legacy institutions will use their cultural authority to delegitimize emerging alternatives. The person with a portfolio and guild membership but no degree will be described as “uncredentialed” — a word that carries connotations of incompleteness and risk. The micro-meritocracy that evaluates people on their work rather than their pedigree will be characterized as “unregulated” or “unaccountable.” The apprenticeship pipeline that produces competent practitioners in two years will be compared unfavorably to the four-year degree on grounds of “breadth” and “well-roundedness” — criteria that conveniently favor the institution making the comparison. The cultural narrative that equates education with university attendance is deeply embedded, and the institutions that benefit from that narrative have every incentive to reinforce it.
How effectively can legacy institutions suppress or co-opt the emerging alternatives? How long can the credential monopoly hold in the face of technological pressure? At what point does the gap between the credential’s cost and its value become wide enough that the coordination problem solves itself — that employers and students simultaneously abandon the old system not because anyone organized the transition but because the fiction became too expensive to maintain? These are questions about institutional power, cultural inertia, and the speed of social change, and they do not have determinate answers. The structural pressures are real. So is the institutional capacity to resist them. The outcome depends on the relative strength of forces that are genuinely difficult to predict.
The Meta-Failure: Optimism as Ideology
There is a failure mode that encompasses all the others, and it is worth naming explicitly: the tendency to treat structural analysis as implicit prescription, and to assume that because a transition is structurally legible, it will be humanely managed.
Nothing in the preceding analysis guarantees that the decomposition of the university will produce a better system. It may produce a more efficient system that is also more brutal — one that sorts people with greater accuracy but provides no safety net for those who are sorted to the bottom. It may produce a system that is more honest about what it measures but less generous in what it provides — one that replaces the university’s inefficient bundling of education, childcare, socialization, and class reproduction with a set of lean, optimized alternatives that serve each function well in isolation but leave gaps where the bundle’s redundancies once provided unexpected support. The student who went to college and learned nothing of academic value but formed the friendships that sustained them through a difficult decade — that person was served by the bundle’s inefficiency. The optimized system has no place for that kind of accidental grace.
The technology sector’s default narrative — disruption as progress, unbundling as liberation, efficiency as justice — is not an analysis. It is an ideology. It assumes that the market’s solution to institutional dysfunction will be superior to the institution’s solution, and it ignores the historical record, which is considerably more mixed. The market’s disruption of newspapers produced a more efficient information ecosystem and also produced an epistemic crisis. The market’s disruption of retail produced lower prices and also produced warehouse labor conditions that would have been illegal in the factories the retail economy replaced. The market’s disruption of the music industry produced universal access and also produced an economy in which the median musician earns less than the median barista. Efficiency is not justice. Disruption is not progress. Unbundling is not liberation. These are outcomes that must be evaluated on their specific terms, not assumed to be beneficial because they are structurally inevitable.
The honest position is this: the educational system as currently constituted is failing, and the structural pressures described in this essay are real and will produce significant change. Whether that change constitutes improvement depends entirely on choices that have not yet been made — choices about who bears the cost of transition, who builds the bridges between old and new systems, who ensures that the functions the old system served badly but did serve are not simply abandoned in the name of efficiency. These are political choices, not technological inevitabilities. And they will be made — or defaulted on — by people, not by structural forces.
The questions in this section do not have answers yet. That is the point. An analysis that identifies structural pressures without acknowledging the places where those pressures could produce catastrophic outcomes is not honest — it is salesmanship. The transition described in this essay will happen. Whether it happens well is an open question, and the openness of that question is not a weakness in the analysis. It is the most important thing the analysis has to say.
8. Conclusion: What Education Was Always Actually For
Begin at the beginning. In the mid-nineteenth century, a set of institutions was designed to solve a specific problem: how to take a large, heterogeneous, mostly agrarian population and transform it into a workforce capable of operating an industrial economy. The solution was compulsory schooling — standardized, age-batched, schedule-driven, authority-centered — and it worked. It produced literate, numerate, socialized citizens at a scale no previous civilization had achieved. The design was honest about its purpose, even if the rhetoric that surrounded it was not. The factory needed workers who could follow instructions, tolerate monotony, and show up on time. The school produced them.
Over the following century, a second layer was added. The university took the sorted output of the school system and sorted it again — more finely, more expensively, and with considerably more pretension. It bundled together credentialing, networking, professional socialization, research, cultural reproduction, and the coming-of-age transition into a single residential package, and it charged accordingly. The bundle made sense in a world where knowledge was scarce, where social networks required physical co-presence, and where the economy changed slowly enough that front-loaded training could last a career. The university’s real product was never the education. It was the sorting — the stamping of individuals with markers of class, competence, and institutional affiliation that employers could read at a glance. The education was the cover story. The selection was the product.
Both layers — the school and the university — operated on a shared assumption so fundamental it was rarely articulated: that knowledge is scarce, that access to knowledge is valuable, and that the ability to demonstrate knowledge is a reliable proxy for the ability to use it. This assumption underwrote the entire system. It justified the credential. It justified the tuition. It justified the four-year residential program, the lecture hall, the examination, the diploma. If knowledge is scarce, then the institution that controls access to knowledge controls access to economic life. If the demonstration of knowledge is a reliable signal, then the institution that certifies the demonstration is performing a genuine sorting function. The whole edifice rests on these two pillars.
AI has knocked out both pillars.
Knowledge is no longer scarce. It is ambient, ubiquitous, and free. The lecture that once required a seat in a specific room is available on any device. The textbook that once cost $150 is summarized, explained, and contextualized by a model that will answer follow-up questions at 3 a.m. with infinite patience. The research paper locked behind a paywall is synthesized on demand. The informational content of a university education — the thing the institution was nominally selling — has a market price approaching zero.
The demonstration of knowledge is no longer a reliable signal. When anyone with access to a large language model can produce a competent legal brief, a structured policy analysis, a well-reasoned essay, or a persuasive grant proposal, the ability to produce those artifacts no longer distinguishes the person who understands the material from the person who typed a prompt. The performance layer — the visible, assessable, legible markers that institutions spent decades using as proxies for genuine understanding — has been automated. The signal has become noise. The proxy has collapsed.
What remains, after the scarcity of knowledge and the reliability of its performance have both been eliminated, is the set of things that were always actually valuable but that the system was never designed to measure: judgment, taste, the ability to know what matters, the capacity to act under genuine uncertainty, the wisdom to distinguish between a plausible answer and a true one. These capacities are real. They are human. They are, for now, beyond AI’s reach. And they are precisely the capacities that the industrial-era school was designed to suppress and that the university was designed to simulate rather than cultivate. The factory model trained compliance, not judgment. The credential certified performance, not understanding. The system was optimized for legibility at scale, and legibility at scale is the enemy of everything subtle, contextual, and genuinely difficult to evaluate.
The emerging institutional landscape reflects this inversion. Where the old system was large, anonymous, and organized around the mass production of standardized credentials, the new system is small, high-trust, and organized around the direct observation of actual work. Guilds replace alumni networks. Apprenticeships replace lecture halls. Contribution graphs replace diplomas. Peer attestation within communities of practice replaces the HR filter’s binary check on institutional affiliation. The unit of organization shrinks from the university of forty thousand to the cohort of forty, because forty is roughly the number of people who can maintain the kind of mutual knowledge — I have seen your work, I know what you can do, I will vouch for you — that makes honest evaluation possible.
This is not a utopian vision. It is a structural prediction with severe failure modes. The transition from credential-based sorting to reputation-based sorting threatens to re-aristocratize opportunity, because building a reputation requires precisely the kind of financial cushion that debt-financed credentials were designed to provide. The verification of real work in a world of synthetic output remains an unsolved problem. The cultivation of taste and cross-domain judgment outside the academy has no proven institutional form. Legacy institutions will fight the transition with every tool at their disposal — regulatory capture, cultural delegitimization, the sheer inertia of a system that employs millions and holds trillions in assets. The bridge between the old world and the new one has not been built, and the people who most need it have the least power to demand it.
But the direction is clear, even if the destination is not. The bundle is coming apart. Each function the university served is migrating toward whatever institutional form serves it most efficiently and most honestly. Credentialing becomes continuous and contribution-based. Networking becomes functional and practice-based. Professional socialization becomes apprenticeship. Research becomes hybrid and distributed. The coming-of-age transition becomes intentional and unbundled from the six-figure price tag. What remains of the university after this decomposition is not nothing — large-scale research infrastructure, the preservation of cultural heritage, the maintenance of intellectual communities too small to sustain themselves in the market — but it is much smaller, much less expensive, and much more honest about what it is.
The core insight is this: AI does not kill educational institutions. It disaggregates them. It strips away the bundling logic that held together functions which were never logically unified, and it forces each function to justify itself independently rather than hiding behind the others. The credential can no longer hide behind the network. The network can no longer hide behind the research. The research can no longer hide behind the coming-of-age experience. Each component must stand on its own merits, serve its actual purpose, and be honest about its actual cost.
This disaggregation is, in the end, a confrontation with a question that the educational system has spent a century and a half avoiding: what is education actually for? The industrial-era answer — producing compliant workers — was honest but is now obsolete. The university-era answer — certifying knowledge acquisition — was always partially dishonest and is now fully untenable. The emerging answer is harder to articulate, because it points toward capacities that resist measurement, institutions that resist scaling, and values that resist commodification. Education, stripped of its industrial packaging and its credentialing pretensions, turns out to be about the cultivation of judgment in the presence of uncertainty — the slow, difficult, irreducibly human process of learning to think well about things that matter, in the company of people who can tell the difference.
That process does not require a campus, a football team, a $200,000 price tag, or a four-year residential program. It requires a small number of people who already possess the judgment in question, a small number of people who are genuinely trying to develop it, a structure that brings them together with enough intensity and duration for the transmission to occur, and an honest accounting of what is being offered and what it costs. Everything else — the administrative apparatus, the credential monopoly, the real estate portfolio, the alumni fundraising machine, the elaborate fiction that all of this is primarily about learning — is overhead. It was always overhead. We just couldn’t see it, because the bundle obscured the components, and the components obscured the purpose.
AI has made the overhead visible. What we do with that visibility — whether we build something more honest in the space the old institutions are vacating, or simply replace one set of fictions with another — is not a technological question. It is a human one. The structural pressures will do the disaggregating. The question of whether what emerges is better or merely different — more just or merely more efficient, more honest or merely more legible — belongs to us. It always did. The institutions just made it easy to pretend otherwise.