Brad DeLong identifies a paradox that will define the next decade of higher education: the very technology that threatens to make student work worthless is the catalyst that will finally force universities to abandon their inefficient, industrial-era teaching models. He argues that the arrival of artificial intelligence is not a crisis of cheating, but a "problemtunity" to resurrect the Socratic method that the modern research university abandoned for the sake of budget convenience.
The Economics of Attention
DeLong begins with a stark admission about the structural incentives of the modern university. He notes that the current system was never designed for optimal learning, but rather for economizing on faculty attention. "We built the modern large-lecture university course as a machine for economizing on faculty attention," he writes. This is a damning indictment of a system that prioritized scale over substance, treating education as a commodity to be processed rather than a skill to be cultivated.
The disruption is immediate. DeLong cites a graph shared by Paul Novosad and Derek Thompson showing that while AI allows students to complete homework faster and achieve higher grades, they are subsequently "crushed on the exam." The gap between the product (the essay) and the process (the thinking) has widened to the point of collapse. As Derek Thompson observes, without a shift to in-person assessment, "AI is going to turn a whole lot of education into the informational equivalent of 'wow I [paid a guy, who] squatted 200 lbs at the gym yesterday, new personal record!'"
This comparison is potent because it highlights the difference between output and capability. A student can generate a perfect essay, but if they cannot explain the causal mechanisms behind it, they have gained no human capital. DeLong argues that the traditional model of grading artifacts is broken because "the feedback loop that would both enable and force student improvement was not closed."
We dressed up a budget constraint we had constructed for our convenience as a philosophy of education.
Critics might argue that mandating one-on-one meetings for every student is logistically impossible in large public universities. DeLong acknowledges the math is tight—75 students, one TA, and ten assignments could consume a full workweek—but he suggests a randomized sampling method could make it feasible. The real barrier, he implies, is not the math, but the willingness to stop treating teaching as a low-cost assembly line.
The Socratic Reset
The core of DeLong's proposal is a radical shift in assessment: every piece of written work must be followed by a ten-minute, one-on-one dialogue. This is not a defensive measure to catch cheaters; it is a "commitment device" to ensure the student owns the ideas they submit. "When I sit across from a student and we argue for ten minutes about the causal story in her paper... I am measuring the stock of judgment directly, at the source," DeLong writes.
This approach reframes the professor not as a "border guard" hunting for smugglers, but as a mentor verifying the chain of evidence. It forces the student to be a different writer before they even hit submit, knowing they must defend their choices in person. "A student who knows that next week she will have to explain this to the professor or the TA is, this week, a different writer," he notes.
This logic aligns with the historical Socratic method, where knowledge is not transmitted but extracted through rigorous questioning. By requiring students to articulate their reasoning, the system ensures that the "judgment" remains a human skill, distinct from the "prose" which AI can now generate cheaply.
The tutorial—Oxbridge's expensive glory, which the American research university abandoned as a luxury it could not scale—turns out to be the assessment mode that a world of cheap text forces back upon us.
The implication is profound: the "luxury" of personalized feedback was actually the minimum viable product for genuine education. The mass-lecture model was the deviation, a cost-cutting measure that masqueraded as efficiency.
The Asymmetry of Text and Judgment
DeLong expands the argument by integrating the work of Johan Fourie, who posits that while text has become cheap, judgment has not. Fourie warns that if researchers use AI to draft first, they risk "stopping [thinking] earlier—too early, in fact," accepting a plausible default rather than pursuing a novel insight.
DeLong synthesizes this with a historical reference to Deirdre McCloskey's 1985 critique of the separation of content and expression. McCloskey argued that "scholarship cannot be written as the sum of two functions," because the act of writing is often the act of discovering the flaw in the argument. DeLong agrees, noting that the "blank page" was a necessary constraint that forced the struggle of thought.
Every piece of writing produces two things: the manuscript you publish now and the judgement you will develop.
The danger, as Fourie and DeLong both highlight, is that AI allows the "unaided capability" to atrophy while "assisted output" rises. The gap in performance narrows, but the gap in true understanding widens. The solution is not to ban the tool, but to change the sequence: own account → model for language and criticism → independent verification.
This distinction is crucial for the future of research. If a historian writes their own account first, the AI can refine the prose. If they read the AI's account first, the "awkward document in the archive" that might have challenged their assumptions may never disturb the familiar story. The tool changes the default, and the default changes the outcome.
Bottom Line
DeLong's most compelling argument is that AI will not destroy higher education, but rather strip away its pretenses, forcing a return to the Socratic dialogue that was discarded for the sake of scale. The strongest part of this analysis is the reframing of one-on-one assessment not as a burden, but as the only way to measure the "stock of judgment" in an age of synthetic text. However, the biggest vulnerability remains the principal-agent problem: while individual professors like DeLong may be willing to trade a workweek of their time for this model, the institutional machinery of the university is unlikely to restructure its budget or class sizes to support it without significant external pressure. The technology is ready; the will to reorganize the labor of teaching is the missing variable.