A few weeks ago I came across a set of stories about how NYU, the University of Chicago, the University of Sydney, and Cal State Maritime Academy are each adjusting to AI in the classroom. I kept coming back to it, and wanted to work through what I think is actually interesting here.

None of these schools are still arguing about whether to ban AI. When a technology delivers real productivity, blocking it doesn’t really work — students use it, the industry they’re headed into already runs on it, and a ban mostly produces students who comply on paper and use it anyway behind closed doors. What these schools are doing instead is moving the place where they check whether a student actually understands anything.

Sydney runs two tracks at once. One assumes students will use AI to produce their coursework and focuses on building the practical skill the industry wants. The other requires a timed oral exam after the paper is submitted, where the student has to explain the reasoning behind it on the spot. If they can’t explain it, the paper wasn’t really theirs.

Cal State Maritime’s approach is more direct. A math professor moved his lectures online and reserved classroom time for something AI can’t do: students working problems at a whiteboard while he watches their thinking unfold step by step. He can’t stop anyone from using AI at home, but he can move the moment of verification somewhere AI has no access to.

NYU’s business school built something called Viva, an AI oral examiner. Students can use any AI tool while working on a project, but afterward they face a round of questioning that gets harder the better they answer — students who’ve been through it say the ceiling keeps rising as long as you keep answering well. The design doesn’t treat AI as something to keep out. It uses AI itself to push students toward a deeper level of understanding.

Chicago Law took a different route. Its first-year core course is piloting a full ban on electronic devices, on the theory that struggling through something difficult is itself how professional judgment gets built. This isn’t a new way of checking understanding — it’s an attempt to hold onto the old one, purely through institutional will.

Looking at the four together, what’s more interesting than the ban-or-don’t-ban question is that each school is answering the same one: if you can’t keep AI out, where does the real test move to? Sydney and NYU land on a similar answer — that a person’s real capability still comes down to whether they can operate independently of AI, whether judgment gets internalized enough that, under pressure, they can walk back through the whole reasoning themselves. Human and AI performance stay two separate, separately testable things. Cal State Maritime sits in between, testing real-time thinking rather than after-the-fact explanation. Chicago didn’t move anywhere. It’s betting the old ground is still worth holding.

How a university tests students doesn’t happen in isolation from what companies actually want. NYU’s Viva design fits corporate taste especially well, because it’s fundamentally an outcomes-based evaluation — whatever tools you used along the way, what counts is how hard a problem you actually solved. I can imagine hiring moving in a similar direction: AI interview tools built to find the people who can still solve the hardest problems with AI assistance. That only works on one condition — different people’s AI-assisted results still have to show a real spread. If everyone using AI ends up producing roughly the same output, this kind of screening stops meaning anything. What’s interesting is that the two sides probably shape each other: what companies want shapes how universities evaluate students, and what universities produce shapes who companies can actually hire. NYU’s design, in a sense, has already pulled that feedback loop forward into the classroom.

This connects to something I keep coming back to in my own thinking about junior engineers. Their training ground is thinning fast because of AI. The old arrangement worked because a junior’s output, relative to their cost, was cheap enough that companies trained them almost as a side effect. Now a senior working with AI tools can often match what a junior produces, and the bar a junior has to clear just to get hired has gone up sharply. What companies dismantled was the training subsidy junior labor used to run on. What these universities are facing is the other half of the same break: the traditional value of a university, transmitting knowledge, was always built on knowledge being expensive to get. Once AI drives that cost toward zero, universities get pushed out of transmitting and into verifying.

I lean toward thinking the dual-track model is the most honest of the four. It doesn’t pretend to have already worked out the answer — it admits the question doesn’t have one yet, and lets different students place different bets. I’m not sure which answer will turn out right: whether operating without AI is just one option among several, or whether it’s the actual foundation that using AI well depends on. I doubt this stays a university question, though. However companies train their juniors, however we end up raising our own kids, sooner or later everyone runs into the same choice.