The Curriculum Half-Life: What AI Is Revealing About Higher Education

63% of college faculty say the graduates they sent into the workforce in spring 2025 were not ready to use AI at work. Those students spent 4 years on campuses where AI was already in everything, which makes that verdict less a judgment on them than a description of what 4 years of college did and did not teach.

Watch a freshman open an AI tutor at 11 p.m. to work through the problem set assigned that afternoon, and the answer arrives in about 8 seconds, correct, with every step laid out in order. The 40 minutes of being stuck, the stretch where a student tries a wrong approach and has to find another one, was the assignment. That part never happens now. The student gets the answer and loses the getting there.

Nobody in that scene is cheating, which is exactly what makes this hard.

Adoption is settled. Judgment is the open question.

Tyton Partners surveyed more than 3,000 students, instructors, and administrators across 750+ U.S. institutions for its Time for Class 2026 report, and found that 61% of students now use AI at least weekly, along with 52% of instructors and 71% of administrators. The argument about whether AI belongs on campus ended a while ago, and it ended without anyone on either side announcing the result.

What has not caught up is everything built around that adoption. Only 32% of institutions have a central AI policy at all, and where one exists, just 22% of faculty believe it works, which means the document meant to govern all of this is one that most instructors have already stopped trusting.

Set that against the Elon University and AAC&U faculty survey published in January 2026, which covered 1,057 faculty and found 95% who fear student overreliance on AI, 90% who expect it to weaken critical thinking, and 74% who worry about the worth of the degree they themselves grant.

So the same faculty are carrying two worries that look like opposites: students lean on AI too heavily, and graduates cannot use it well enough for a first job. Both readings are accurate, and they point at the same missing thing, which is judgment: knowing when the machine's answer is good, and knowing when it is confidently wrong.

The same pattern is running through academic research.

Nature published a study in January 2026 covering 41.3 million papers across the natural sciences, and the individual returns it measured are hard to argue with: scientists working with AI methods published 3.02 times more papers, drew 4.84 times more citations, and reached research leadership 1.37 years sooner. For an individual career, that is decisive.

The same study found that the collective volume of topics under study shrank 4.63%, and that scientists engaged with one another 22% less. More output, aimed at fewer questions, by people comparing notes less often than they used to. AI goes where the data already sits, which means it works the established fields faster and leaves the unmapped ones alone.

A Frontiers survey of roughly 1,600 academics across 111 countries, reported in December 2025, found that more than half had used AI while peer reviewing a manuscript, often against the guidance they had been given. The people who decide whether science is sound are already delegating part of that decision.

The objection was right and the ban still failed.

None of this is the first time a teaching institution met a tool it could not control. In Plato's Phaedrus, Socrates objects to writing on the grounds that students who trust external marks will stop exercising their own memory, and will seem wise without being wise. He was not wrong about the memory, since written text did change what people bothered to remember. Reading then became the foundation of every university that followed, including the one that assigns you the dialogue in which the objection appears.

The objection was accurate and the ban was unenforceable, both at once, and only one of those two facts decided what came next.

Which brings me to the number that never appears in a course catalog: the curriculum half-life. A degree is a 4-year bet, made on a student's behalf, that what gets taught in year 1 still holds in year 4 and for a decade after that. The bet was safe for a long time, and it is getting shorter every year, while the institution absorbs none of the loss when it comes due. The graduate does.

Curricula expire. The habit of relearning does not.

Since 2022 I have been Innovator-in-Residence at Tulane University, and since 2025 I have advised the California State University system on AI adoption and governance: 22 campuses, 460,000 students, and a $17M ChatGPT Edu deployment that put the same tools in front of a first-generation student in Bakersfield and a graduate researcher in San Diego. CSU made that call at scale, on its own initiative, at a moment when very few systems anywhere were willing to. Whatever anyone concludes about the contract terms, the access question got answered for half a million people.

Access was the easier half.

The harder half is what a student does with 4 years now that the answer to nearly any assigned question is 8 seconds away. Multiply that 11 p.m. problem set across a few million students and it settles two things no faculty senate will ever vote on: whether a graduate can tell a good answer from a plausible one, and whether a family's tuition bought a credential or bought a capacity.

I spent 20 years at Cisco, most of it in innovation strategy, where I shaped a $1.1 billion innovation portfolio that generated more than $400 million in revenue. The fastest-expiring asset we owned in all that time was our own assumption that the skills which produced the last result would produce the next one.

I don't have a settled answer here, and neither does anyone I work with inside these institutions. There is one question worth carrying into a curriculum committee, a dissertation defense, or a campus tour, and it costs nothing to ask: what does this program teach that a student will still be able to use after the tools we taught it on are retired?

If you are a student or a parent rather than a provost, that question belongs to you too, and the person to ask is the admissions officer, politely and directly. Programs worth the tuition can answer it. And if you see this differently, I'm easy to find.

Here is what makes Alex a credible voice on this topic: Alex has been Innovator-in-Residence at Tulane University since 2022 and has advised the California State University system on AI adoption and governance since 2025, work that puts him inside the meetings where colleges decide what AI access means for 4 years of coursework.

For more, book a conversation →