Ask what an exam actually certifies and the answer is narrower than the ceremony suggests: that on a given day, a person retrieved stored information under time pressure without external aids. For a century that was a reasonable proxy for professional readiness, because storage and retrieval were scarce human capacities and the workplace resembled the exam room closely enough. Then retrieval became a utility. When any knowledge worker can summon the substance of most curricula in seconds, a credential certifying internalized recall is certifying a capability the market has begun repricing toward zero.
The repricing is no longer hypothetical, and its first victims are exactly the people whose main asset is fresh credentialed knowledge. Research from the Stanford Digital Economy Lab (Brynjolfsson, Chandar and Chen, 2025), using large-scale payroll data, found that early-career workers aged 22 to 25 in the occupations most exposed to AI experienced a relative employment decline of about 16 percent, while experienced workers in the same occupations held steady. The study's sharpest distinction is between work AI automates and work it augments. Entry-level knowledge work is disproportionately the automatable kind, because entry-level knowledge work is, to a first approximation, the application of recently memorized material to well-specified problems. That is what the credential certified. That is what the machine now does.
The demand side is shifting at matching speed. The World Economic Forum's Future of Jobs Report 2025 projects that nearly 40 percent of workers' skills will change by 2030, with 63 percent of employers already naming the skills gap their top barrier. A credential system whose products describe a graduate's stock of knowledge is issuing statements about an asset with an accelerating depreciation schedule.
What was always upstream of recall
None of this means knowledge stopped mattering; it means the certifiable layer of it stopped being the scarce layer. What AI exposes is a distinction educators always knew and assessment systems mostly ignored: between having information and knowing what to do when the information runs out. Framing an ambiguous problem. Deciding which of five plausible machine outputs is subtly wrong. Taking responsibility for a result in front of people affected by it. Carrying a project from intention through friction to done. These capacities were never well measured by closed-book exams, and they are precisely what remains expensive after retrieval becomes free.
The awkward institutional truth is that memorization persisted as the backbone of credentialing not because anyone believed it was the essence of competence, but because it was cheap to test at scale. Multiple-choice questions grade themselves. Judgment does not. The credential system optimized for what was assessable, then the labor market treated the assessable as the valuable, and AI has now called the bluff on the whole chain.
Schools that bet against recall early
There have long been working prototypes of the alternative. Ricardo Semler has spent two decades backing schools, some 3,000 children across six countries, where students choose what to learn and adults act as tutors rather than bosses: no rows, no bells, no grades for rote recall, just a project, a tutor, and a deadline, 'which is what work looks like too,' as he puts it. His larger jab was that the school with rows of desks was a nineteenth-century factory preparing children for factories that no longer exist. The AI moment gives the jab new teeth: the exam hall now prepares students for a competition with software, on the software's home field.
A project with a deadline, by contrast, assesses the post-retrieval capacities natively: the student must scope, choose, persist, and present something that works. It is assessment by evidence of work rather than by evidence of storage, the same substitution the hiring world is being forced toward.
The credential that survives
What replaces the memorization credential is not credential-free chaos but a shift in what proof is made of: portfolios of completed work, records of judgment exercised under real stakes, testimony from people who watched it happen. These are harder to grade and harder to fake, which is the entire point. Verification will get its own tools; the primary evidence must change first.
The exam certified what the mind could hold. The market now pays for what the mind can do with what any mind can hold. Institutions that keep certifying storage will keep issuing, at rising volume and falling value, beautifully sealed receipts for a commodity in surplus.
Sources
- Brynjolfsson, Chandar & Chen (2025). Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of AI (Stanford Digital Economy Lab). Early-career workers (22-25) in the most AI-exposed occupations saw ~16% relative employment decline; losses concentrate where AI automates rather than augments. https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/
- World Economic Forum (2025). Future of Jobs Report 2025. ~40% of workers' skills expected to change by 2030; 63% of employers cite the skills gap as the main barrier to transformation. https://www.weforum.org/publications/the-future-of-jobs-report-2025/