Three of the most powerful people in business made three contradictory claims about AI and employment within the same ten-week stretch this spring. All three were right. Here's the data that explains why - and the one number nobody's talking about.
This spring, BlackRock CEO Larry Fink told investors the Class of 2026 could see the worst graduate job market in over a decade, "even without a recession." Around the same time, OpenAI's Sam Altman told a podcast audience this is "the most exciting time to be starting out one's career, maybe ever." And on May 29, Apollo Global Management's chief economist Torsten Sløk published a memo with a title engineered to go viral: zero evidence of job losses because of AI.
A BlackRock billionaire, an OpenAI billionaire, and a Wall Street chief economist. Same three months. Opposite conclusions. None of them lying.
The labor market didn't get destroyed by AI this year. It got sorted. Which headline you believe depends entirely on which side of the sort you're standing on - and the data from the last 90 days draws that line with unusual precision.
The rung that disappeared
Start with the casualty everyone agrees on: the first job. The unemployment rate for U.S. college graduates aged 22 to 27 sits at 5.6%, according to the Federal Reserve Bank of New York — a level not seen outside of the pandemic since 2013. Handshake's Class of 2026 Network Trends report shows entry-level postings down 2% year-over-year and 12% below pre-pandemic levels. Monster's survey of this year's graduating class found 90% worried AI or automation will take their first job, up from 64% just last year.
Goldman Sachs has tried to put a number on the bleeding: somewhere between 11,000 and 16,000 AI-linked job cuts a month in the U.S., depending on which of the bank's two 2026 estimates you use — concentrated overwhelmingly among workers under 30. Anthropic's own CEO, Dario Amodei, has been saying for over a year that AI could eliminate up to half of entry-level white-collar roles. A growing number of this year's graduates aren't waiting to find out: ZipRecruiter's 2026 Graduate Report found 38% considering starting their own business, 32.5% planning to freelance or gig their way in, and 11% heading straight for the skilled trades instead.
That's the Fink headline. It's real. It's just not the whole picture.
The lane nobody's watching
PwC's 2026 Global AI Jobs Barometer - released June 15, built from more than a billion job postings worldwide - found something that doesn't fit the doom narrative at all. Jobs requiring real AI skills are growing 69%, nearly eight times faster than the 9% growth of the overall job market. The wage premium for those skills has climbed to 62%. The number of postings explicitly requiring AI skills has roughly doubled since 2024.
PwC calls it a two-track labor market. Call it the Expert Lane and the Easy Lane. In the Expert Lane — radiologists, recruiters, financial analysts - AI handles the routine work and what's left over demands more human judgment than ever. These roles are growing twice as fast and paying 42% more than the Easy Lane, where AI just makes the job simpler for a non-expert to do - IT service desks, medical secretaries, call center supervision.
Here's the detail that should unsettle anyone clinging to "AI is killing entry-level jobs" as a clean story: the entry-level roles that are AI-exposed and still hiring are seven times more likely to demand senior-level skills - judgment, leadership - than the ones that aren't. Those specific roles grew 35% since 2019. Ordinary entry-level roles, the ones that don't ask for judgment you can't have at 22, fell 10% over the same period. The ladder's bottom rung isn't gone. It just got moved up to where only people who already have experience can reach it.
The confound nobody wants to admit
Before this turns into another tidy AI-villain story, the rigor demands a pause. Sløk's "zero evidence" memo leans on ADP data showing private payrolls added roughly 110,000 jobs in April, and argues this is Jevons paradox in motion: cheaper AI generates more spending on AI implementation, data centers, and specialists — not less employment overall.
Separately, researchers Lambert and Schindler noticed something the AI-blame narrative had skipped past entirely: the occupations most exposed to generative AI and the occupations most exposed to remote work are, statistically, almost the same occupations — white-collar, computer-heavy, easy to offshore. The entry-level hiring collapse lines up almost exactly with ChatGPT's late-2022 arrival. It also lines up with the post-pandemic remote-work boom that made junior roles easier to outsource globally. Untangling which force did the damage is harder than any headline admits.
The UK data adds its own confound. NIESR research attributes roughly 7% of the rising cost of hiring entry-level UK workers to increases in National Insurance contributions, minimum wage, and employment-rights reforms - not AI at all. And per the British Chambers of Commerce, AI adoption among UK SMEs has more than doubled, from 25% to 54%, in eighteen months — meaning even where AI usage is exploding, it's arriving alongside several other cost and policy shocks hitting the same junior roles at the same time.
This is the part that doesn't make it into the viral version: AI is a real force in the entry-level collapse, but it is not the only one, and nobody serious has fully separated its share from the others yet.
Where the money actually went
If companies were truly slashing headcount because of AI, you'd expect to see it in the replacement data. You don't - not at scale. ManpowerGroup's 2026 Talent Shortage Survey of 39,000 employers across 41 countries found that only 10% of UK employers are using AI or automation to directly replace headcount. What's actually happening is something closer to a freeze-and-redirect: companies are slowing full-time, junior hiring and pouring the savings into renting senior expertise instead.
Upwork's 2026 In-Demand Skills report makes the redirection explicit: 77% of business leaders say AI is increasing their need for specialized, fractional talent rather than traditional full-time roles. AI-related freelance work grew 109% year-over-year, with AI integration work up 178% and AI data annotation up 154%. Specialized AI freelancers are now commanding 25-60% higher rates than general practitioners in the same field - and the premium is widening, not shrinking.
Put the two data sets side by side and the real 2026 story comes into focus. It was never "AI is taking jobs." It's that AI has made deep, specific expertise the only scarce resource left — and companies have figured out they don't need to own that resource on a permanent payroll to access it. The job didn't vanish. The full-time, 9-to-5, build-it-from-a-graduate version of the job did.
The emergence of AI-native talent platforms reflects this shift. Instead of hiring large teams and hoping skills match evolving business needs, organizations are increasingly turning to specialized, on-demand experts. Platforms such as Stynt AI are built around this model, helping companies identify and engage highly skilled professionals for specific outcomes rather than traditional job descriptions. In many ways, these platforms are becoming the infrastructure layer for a labor market that values expertise over headcount.
The $350 million tell
The clearest signal that this is structural, not speculative, came from the company with the most to lose by admitting it. On June 10, Anthropic committed $350 million - a $200 million Economic Futures Research Fund plus $150 million for a program called Claude Corps - to fund independent research and a tiered policy framework for what happens as AI reshapes employment. Buried in the research behind that announcement is a number that cuts against the panic on both sides: measured against real usage rather than theoretical task lists, about 30% of the workforce — cooks, mechanics, bartenders, lifeguards - currently has close to zero AI task exposure, simply because the work is physical and in-person.
A company doesn't put $350 million behind labor-market research because it thinks the effect is mild. But it also doesn't fund a tiered policy framework if it thought the answer was simple, total replacement. The bet implicit in that number is the same one the rest of the data points to: this isn't a story with one ending. It's a redistribution, and redistributions are exactly the kind of thing that needs careful, well-funded measurement instead of a hot take.
So who's right — Fink or Altman?
Both. Fink is describing the Easy Lane and the graduating class trying to enter a ladder whose bottom rung just moved. Altman is describing the Expert Lane, where a 25-year-old with real judgment and AI fluency can now produce work that used to take a team of seniors. Sløk is technically correct that the aggregate jobs number hasn't cratered - because the damage isn't aggregate, it's positional.
The job market in 2026 didn't shrink. It unbundled. Full-time employment is no longer the default unit in which expertise gets delivered - it's becoming one option among several, competing against fractional, on-demand, pay-for-the-outcome arrangements that are growing faster and paying better. For a graduate without judgment to sell yet, that's a crisis. For anyone who already has it, untethered from a single employer's payroll, it might genuinely be the best time in history to be good at something specific.
The ladder didn't disappear. It just stopped being free to climb.
Top comments (0)