A front-page story in UC Berkeley’s student newspaper, The Daily Cal, has been tearing through the internet this week: failing grades are surging in Berkeley’s computer science courses, and professors are pointing at two culprits — rampant AI use by students, and a precipitous decline in basic math skills. Cue the clucking. Cue the think pieces about the moral rot of the young. Cue the very sincere HN commenters insisting that they learned C on a VT220 and turned out fine.
But pause on that second culprit for a moment. Dwindling math skills.
Students who a decade ago would have ground through linear algebra problem sets are now, apparently, arriving unable to do the work and unwilling to pretend otherwise. The AI tools make it easier to paper over the gaps — submit something plausible, get the points, move on — but the gaps were opening before ChatGPT arrived. What changed?
The labor market did. And Berkeley’s students noticed before Berkeley’s faculty did.
The Pipeline That Closed While Everyone Was Looking at Cheating
For twenty years, the implicit bargain of a Berkeley CS degree went something like this: endure four years of brutal math, data structures, and algorithms, and you walk into a six-figure salary engineering the recommendation systems that keep people scrolling. The math was painful but legible — a toll road with a known destination.
That destination is shrinking. Not the tech industry broadly — the specific, math-heavy software engineering roles that justified the calculus and linear algebra gauntlet. Meta laid off thousands of engineers while explicitly shifting toward AI-augmented workflows. Google’s latest earnings call spent more time on Gemini than on headcount for core engineering. The message filtering down to undergraduates — through internship hunts, through older siblings, through the Slack DMs of recent grads — is that the jobs that rewarded grinding through problem sets are fewer, more competitive, and increasingly skeptical that four years of manual optimization is better than six months of prompt engineering.
A student who watched her cousin, Berkeley CS class of 2022, spend eight months job-hunting in a cooling market is making a different calculation than a student who enrolled in 2018. She’s not lazy. She’s not morally defective. She’s responding to price signals the curriculum doesn’t reflect.
“Two years ago, the advice was still ‘grind LeetCode, get the offer,’” said a teaching assistant in the department who declined to be named because they weren’t authorized to speak to press. “Now half the students I talk to are building AI-wrapper startups on the side. The ones who aren’t? They’re looking at the same job boards I’m looking at.”
What the Moral Panic Overlooks
The easy story here is a morality tale: technology enables sloth, sloth produces ignorance, ignorance produces failure. It is satisfying in the way all decline narratives are satisfying. It confirms that the old ways were better and the young are squandering them.
But the harder story — the one that leaves nobody feeling righteous — is that the students might be making a rational tradeoff. If the market is going to pay you to orchestrate AI agents rather than invert matrices by hand, why spend four years mastering matrix inversion? The professors are grading against a standard that the economy is quietly abandoning.
This is not an argument that math is worthless or that students should cheat. It’s an observation that when the payoff structure changes, behavior changes — and blaming the behavior without examining the payoff is a comfortable evasion. Berkeley’s CS department is running a curriculum designed for a labor market that peaked around 2019. The students are living in 2026 and adjusting accordingly.
The Hard Question Nobody Wants to Ask
The uncomfortable follow-up, for anyone inclined to moralize, is this: who’s actually being irresponsible here? The nineteen-year-old who uses Claude to get through a problem set she’d otherwise fail, because she’s betting her career on skills the problem set doesn’t teach? Or the institution that charges her $15,000 a year in tuition for a credential whose underlying value proposition is eroding, and then blames her when she stops pretending it isn’t?
Berkeley could update its curriculum. It could decide which mathematical competencies remain genuinely load-bearing in an AI-saturated industry and which are vestigial — the academic equivalent of teaching slide rules in 1980. But that would require acknowledging that the faculty’s own expertise is partially depreciating, which no institution does cheerfully.
Instead, we get the failing grades story, which neatly locates the problem in the students’ character rather than the university’s relevance. It’s a trick as old as higher education: when the bridge between classroom and career starts crumbling, blame the pedestrians for finding another route.
The students aren’t the only ones failing a test here.