Day 979 - The Degree That Ate Itself
Enrolments are responding to a nasty reality.
UC Berkeley’s CS department will graduate approximately 350 students in 2027. Two years ago, it graduated over 1,000. That’s not a blip. That’s a collapse.
The numbers behind it are worth sitting with. Job postings for software development roles on Indeed fell 71% between February 2022 and August 2025. US CS graduates now face 6.1% unemployment, higher than philosophy majors (3.2%) and art history graduates (3.0%). The degree that tech executives spent twenty years telling kids was a guaranteed path to six figures now ranks 7th in unemployment among all college majors.
Students noticed. Enrolment in CS programs fell at 62% of universities last year. Deposit volumes for CS degrees dropped more than 25% in a single application cycle. For context: deposits are the leading indicator. What’s in the enrolment data now is the caution of 2024. What those students will find when they graduate in 2027 is not something anyone can model yet.
The same trend is playing out here. In Australia, just 2.9% of incoming university students chose ICT-related degrees in 2026, down from around 9,750 students the year prior to roughly 7,686. The Australian Computer Society called it a “steady decline” with no obvious explanation. In New Zealand, 34% of companies have already slowed entry-level hiring, and 88% expect to do so within three years, the highest reported rate of entry-level job displacement from AI of any country in the IDC survey. New Zealand was also the country reporting the highest rate of roles removed entirely due to AI, at 53%.
The pivot is to AI degrees. UC San Diego launched one; it’s the only UC campus where CS enrolment went up. MIT’s AI and decision-making major is now its second-largest. Dozens of universities are scrambling to repackage. The message to students: same building, new sign out front.
Here’s the problem with that bet. The students fleeing CS for AI degrees are responding to the 2023 labour market, the one where “prompt engineer” was a job title and AI literacy felt like a moat. The 2027 labour market, when those students graduate, is a different question entirely. If the models keep improving at their current rate, the gap between “knows how AI works” and “AI works” will close before the cohort crosses the stage.
There’s a structural irony worth noting. US universities doubled CS output over the last decade, from 52,000 graduates in 2013-14 to 113,000 in 2022-23, partly because the industry kept signalling demand. The same companies doing that signalling then overhired, laid off 260,000 workers in 2023 alone, and began replacing junior roles with AI tooling. The pipeline they filled is now flooding a market they’ve largely closed.
The student debt stays. The jobs didn’t.
Prediction ledger entry: UC Berkeley CS graduating class will remain below 500 in 2028 (currently tracking toward ~350 in 2027). Confidence: 70%. Score on Day 180.
Practical tip
If you’re advising someone choosing a degree right now — a kid, a junior colleague, anyone — the question to ask isn’t “what’s hot?” It’s “what’s the half-life?” Degrees take four years to complete and deliver value over a forty-year career. Any skill with a half-life shorter than the degree itself is a bad investment. CS as traditionally taught is in that category. The question for AI degrees is whether they’re different in kind, or just further along the same curve. Nobody honest can answer that yet. Hedge accordingly.
Sources: UC Berkeley Daily Californian · Computing Research Association CERP Pulse Survey · MARKETview · Federal Reserve Bank of St. Louis / Indeed data · National Center for Education Statistics · ACS Information Age · IDC/Deel AI at Work report (NZ data)