Computer science graduates are ending up waiting tables — and a new Census Bureau study blames AI.
The results are stark: graduates from the most AI-exposed majors saw their initial employment likelihood drop by five percentage points and their full-quarter earnings fall by 13 percent. That's not a rounding error — it's comparable to the earnings losses suffered by millennials who graduated during the Great Recession.
And here's the part that will keep parents up at night: about half of that earnings decline is because these grads are ending up in lower-wage sectors like restaurants and retail. The frappuccino-slurping, plate-carrying compsci graduate is no longer an ironic meme. It's a statistic.
The Numbers That Should Alarm Everyone
The working paper, authored by three US Census Bureau economists, examined college graduates from majors with the highest exposure to AI since ChatGPT's late 2022 debut. The findings paint a concerning picture for anyone who thought AI would only affect "experienced workers" or "routine tasks."
For the top decile of AI-exposed majors — which includes computer science, mathematics, statistics, some engineering disciplines, networking, and yes, even journalism — the early career outlook has deteriorated sharply:
- 5 percentage point decline in likelihood of initial employment
- 13 percent decline in full-quarter initial earnings
- Shifts into lower-wage sectors (restaurants, retail) account for roughly half the earnings drop
Meanwhile, graduates from less AI-exposed fields — nursing, education, many healthcare roles — haven't experienced the same deterioration. The labor market isn't shrinking uniformly; it's being reshaped by AI in ways that disproportionately punish young people who studied the very skills AI can now replicate.
Who's Most Exposed?
The study identifies the usual suspects: computer science, mathematics, statistics, some engineering disciplines, and networking. But it also flags journalism — a field that has been under AI pressure for years, from automated sports recaps to AI-generated news briefs.
This isn't just about coding bots writing basic scripts. The exposure encompasses any role where AI can now perform tasks that were previously the domain of entry-level professionals. Junior developers writing boilerplate code. Analysts generating routine reports. Writers producing formulaic content. These are exactly the jobs that younger workers traditionally entered to build their careers.
The researchers note that other factors could be at play — economic cycles, hiring freezes, uncertainty about AI's trajectory — but the pattern is clear: AI exposure correlates strongly with worse early-career outcomes.
The Pipeline Is Already Changing
One of the most telling findings isn't about jobs — it's about enrollments. The study found that while AI-exposed majors saw rapid graduate growth in the years leading up to ChatGPT, that growth started to level off quickly after its release. Survey data from STEM departments suggests enrollments in some of the most exposed programs are now declining.
This is the market correcting itself — students and families responding to labor signal data faster than any policy intervention could achieve. If AI is making certain careers less attractive, fewer people will pursue them. The question is whether that adjustment happens fast enough to prevent a glut of qualified workers competing for a shrinking pool of traditional roles.
Self-Employment and Grad School: Partial Relief
The researchers found that changes in self-employment and graduate school enrollment account for roughly half of the five-percentage-point employment decline. In other words, some displaced grads are starting their own businesses or retreating to academia — but these aren't solutions for the majority.
The self-employment route is risky, especially in an uncertain economy. Grad school delays entry into the workforce and doesn't guarantee better outcomes if AI is also disrupting academic research and teaching. These are coping mechanisms, not answers.
🔥 Hot Takes
1. The "entry-level paradox" is real and it's dangerous. AI is simultaneously making entry-level work easier to automate while removing the very experiences that help junior workers become senior. You can't learn to code by writing CRUD endpoints if AI writes those endpoints. You can't learn to analyze if AI generates the analysis. The pipeline from junior to senior is clogging because the first rung is being pulled away. This isn't just bad for grads — it's bad for the industries that need to develop their talent.
2. Great Recession comparisons aren't hyperbole — they're data. The study explicitly compares current earnings declines to those experienced by millennials who graduated into the 2008 financial crisis. That's not alarmist language; it's empirical. If we're seeing recession-level damage from a technology shock rather than an economic one, that should recalibrate how policymakers think about AI regulation. This isn't a "disruption" story — it's a labor market trauma story.
3. The enrollment decline is both warning sign and self-correction. Fewer students choosing AI-exposed majors is the market signaling that the value proposition is changing. But it's also a lagging indicator — today's enrollees will graduate into a market that may look very different. And for those already in the pipeline? They're facing the worst of both worlds: training for jobs that are disappearing without a clear alternative path.
The Bottom Line
The Census Bureau study is clear: AI is reshaping the labor market, and young people entering AI-exposed fields are bearing the brunt. The 13 percent earnings decline and five percentage point employment drop aren't abstract numbers — they represent real graduates who expected one trajectory and are facing another.
The researchers caution against long-term doomsday forecasting, noting that historical evidence shows new jobs emerging that we can't yet predict. That's true — but it's also true that the transition period can be brutal for those caught in it. And unlike previous technological transitions, AI doesn't just automate physical labor or routine tasks — it automates cognitive work, including the kind of foundational work that younger professionals have historically used to build their careers.
The frappuccino is no longer a joke. It's a data point. And unless the economy creates new categories of work fast enough to absorb displaced compsci grads, we're going to see a generation of highly educated workers who can't find work in their fields — not because they lack skills, but because those skills are now cheaper to automate than to employ.
Source: The Register | Census Bureau Working Paper