AI's impact on the college graduate job market

The starkest tension in this story isn't between believers and skeptics of AI disruption - it's between two rigorous studies looking at the same market and reaching opposite headlines. A CESifo working paper by Robert Fairlie and Jane Wu, using Census CPS microdata, found recent college graduate unemployment sat at 7.3% in summer 2026, squarely inside the 6.3%-7.8% range recorded in the three prior summers, with no statistically significant relative increase compared to older graduates or non-degree young workers ^[1]. Yet a Stanford Digital Economy Lab analysis led by Erik Brynjolfsson found a 16% decline in early-career employment concentrated specifically in the occupations most exposed to generative AI since late 2022 ^[2]. Both findings can be true simultaneously: an aggregate unemployment rate averages across every major and industry, while the AI-linked shock is landing unevenly, hitting information, finance, insurance, marketing, and back-office roles hardest while other fields absorb graduates as usual. That sector-by-sector unevenness, invisible in a single unemployment percentage, is closer to the real story than either an 'AI apocalypse' or 'nothing to see here' headline suggests. The same fault line played out in public in real time: after data circulated showing new-grad hiring up 5.6% year-over-year, investor David Sacks pointed to it on X as evidence the AI-job-apocalypse narrative was overblown, while The Economist countered on the same platform with alumni-survey data showing full-time employment among the most AI-exposed graduating cohorts fell from nearly 70% to 55% over the three years since ChatGPT's 2022 release - the aggregate-versus-concentrated argument, replayed tweet for tweet.

Long before any human recruiter reads a resume, an estimated 90% of U.S. employers now route entry-level applications through AI-powered applicant tracking systems that score and rank candidates, with AI-driven ATS workflows cutting time-to-hire roughly in half ^[3]. For new graduates with no work history to differentiate them, this turns hiring into a keyword-matching contest against a black box, and it is already drawing legal scrutiny: a January 2026 California lawsuit against AI-hiring platform Eightfold alleges its screening tools produce discriminatory rejection outcomes ^[3]. At the same time, employer appetite for candidates who can operate this new machinery is climbing fast - NACE's Job Outlook 2026 Spring Update found the share of entry-level postings requiring AI skills nearly tripled since fall 2025, now covering more than a third of listings ^[4], while resume-side data shows AI-skill mentions among graduating seniors have doubled since the Class of 2022 ^[5]. The gate is tightening and the price of admission is rising at the same time. That opacity fuels genuine disagreement even among people close to the technology: in a Reddit thread about AI hiring frustration, a self-identified applicant-tracking-system industry worker pushed back that none of their customers automate outright rejections and that a human always reviews a file before anyone is turned down - a claim job seekers in the same thread met with skepticism, describing the process as an inscrutable black box regardless of who is technically pulling the trigger.

The fight over what is actually causing entry-level cuts is at least as sharp as the fight over the data itself. BlackRock CEO Larry Fink has warned that AI is disrupting the traditional college-to-white-collar pathway and could push Class of 2026 unemployment to its highest level in years even without a recession ^[6]. Marc Andreessen rejects that framing outright, arguing large companies overstaffed by 25% to 75% during the pandemic and are now using AI as a convenient excuse for cuts that predate AI's actual task capability ^[7]. Andreessen goes further, predicting AI-driven productivity gains will eventually trigger a hiring boom rather than a bust ^[8]. Meanwhile Anthropic's own leadership has warned entry-level white-collar jobs could be halved within one to five years, a claim Yale's Jeffrey Sonnenfeld reframes as understating the problem: the damage isn't visible layoffs but the entry-level roles that quietly never get created at all ^[2]. That framing reached a mainstream audience too: a CBS Evening News segment drawing on Oxford Economics research told viewers new-grad unemployment had exceeded the national rate for the first time, pinning the cause on employers reassigning the repetitive, cognitively routine tasks that used to be a junior hire's on-ramp.

Employer demand for AI fluency at the entry level isn't just rising, it's compounding: NACE found AI-skill requirements in postings nearly tripled since fall 2025, with 28% of employers specifically recruiting early-career talent who can use AI and nearly 60% assigning interns AI-tool projects ^[4]. Handshake's data shows students responding, but unevenly - AI-skill mentions on resumes have doubled since the Class of 2022, and AI keywords now appear in a meaningfully larger share of full-time entry-level postings and internships than a year prior ^[5], with Handshake itself building an AI resume tool to help students keep pace ^[9]. Yet NACE also found nearly a third of graduating seniors think AI skills will matter little to their careers, and just over half aren't building them - even as employer demand triples. That mismatch, not AI's raw capability, may be doing as much to sort winners from losers in this hiring cycle as anything else.

Even where graduates do find work, the quality of that work is eroding in ways a single unemployment rate cannot capture: underemployment among recent college graduates reached almost 43% by the end of 2025, up nearly four points from the year before ^[10]. That gap between 'employed' and 'employed in a job that uses the degree' echoes a pointed observation from one Reddit discussion of the trend: a commenter there described the shift not as elimination but consolidation - work that once took a team of five to ten people now getting done by two or three. It's an anecdotal framing, not a data point, but it reconciles the CESifo team's own caution that summer 2026 is an initial test rather than a final verdict ^[1]with the sense among some current job seekers that something has quietly narrowed even where the official numbers still look fine.