What Is Actually Happening to Jobs? Separating AI Hype from Reality

# ai# technology

The AI Prism

Originally published on The AI Prism Here’s the uncomfortable truth about the AI jobs debate: the...

Originally published on The AI Prism

Here’s the uncomfortable truth about the AI jobs debate: the loudest voices have already moved on, and the data is now telling a far more interesting story than either the doomsayers or the dismissives predicted.

In May 2025, Anthropic CEO Dario Amodei predicted AI could wipe out half of all entry-level jobs within one to five years. By May 2026, OpenAI’s Sam Altman was saying he doubts “we’re going to have the kind of jobs apocalypse that some of the companies in our space advocate or talk about.” That is a spectacular reversal in 12 months — and it tracks with what the numbers actually show.

This month, a Stanford SIEPR policy brief — written by economists including the former Commissioner of the Bureau of Labor Statistics — landed on Hacker News and drew 300+ points and 377 comments. Its title could be ours: “What is really happening to jobs? Separating AI hype from reality.”

We dug into the brief, the underlying datasets, and the labor market numbers behind it. Here is what is actually happening — which roles are growing, which are shrinking, and which are simply being rewritten.

The Doomsayers Are Quietly Walking It Back

Start with the people who set the terms of the debate. Amodei’s 2025 prediction — half of entry-level jobs gone in one to five years — was the ceiling of the apocalypse narrative. He followed it in January 2026 by calling AI a potential “general labor substitute for humans,” and warned of a world stuck on “hypergrowth, hyper-inequality.”

Then the tone shifted. A Fortune report in May 2026 documented both Altman and Amodei walking back their predictions, and the WSJ reported Big Tech had “suddenly flipped” on the jobs wipeout scenario. Even the Guardian ran the headline “The AI jobs apocalypse probably isn’t coming anytime soon” in July 2026.

The about-face isn’t purely rhetorical. Anthropic’s own research arm published a labor market analysis in March 2026 finding “no systematic increase in unemployment for highly exposed workers since late 2022” — and noting that Claude currently covers just 33% of tasks in the computer and math category, even though it could theoretically handle nearly 100%.

MIT economist David Autor, one of the most cited labor scholars in the field, put it bluntly: “A lot of people have noticed that the world is not changing as fast as they predicted.”

The Macro Data: No AI Recession — Yet

Here’s the headline number from the Stanford brief: since 2022, unemployment among the most AI-exposed workers has risen 0.77 percentage points — while unemployment among the least exposed workers rose 0.85 points. In other words, the workers most at risk from AI are faring slightly better than everyone else. That is not the signature of an AI-driven jobs crisis; it’s the signature of a broadly softening economy.

The same pattern shows up in the actual employment counts. BLS data for computer systems design — the sector that should be ground zero for AI displacement — shows employment essentially flat since ChatGPT launched: 6.71 million workers in November 2022, 6.67 million in June 2026, a decline of roughly 0.7% over 3.5 years. During that same window, the series peaked at 6.73 million in late 2025 before drifting down. Flat is not collapse.

Apollo chief economist Torsten Slok ran the same check in June 2026: if AI were triggering a jobs crisis, job openings would be collapsing. Instead, the ratio of openings to unemployed workers climbed back above 1.0, and May’s jobs report showed nonfarm payrolls up 172,000. “There are no signs of workers being replaced by ChatGPT,” Slok concluded.

LinkedIn’s own economic graph — a billion members’ worth of hiring data — agrees. Chief Global Affairs Officer Blake Lawit confirmed in April 2026 that hiring is down about 20% since 2022, but explicitly pushed back on AI as the cause: “We’ve looked — and honestly, we haven’t seen it.” His attribution: interest rates.

Even the firms that adopted enterprise AI are hiring, not firing. The Stanford brief cites research showing employment at AI-adopting firms grew 10% in the two years after adoption.

The Graduate Squeeze Is the One Real Signal

Now for the part that should worry you: new graduate unemployment hit 5.6% in early 2026, up 1.6 percentage points in three years. That is the single clearest labor market change of the AI era, and it’s the one place where the data and the doom narrative actually line up.

Stanford Digital Economy Lab research (Brynjolfsson, Chandar, and Chen), using ADP payroll data, found employment among early-career workers in AI-exposed occupations — software developers and customer service representatives — declined noticeably after ChatGPT’s launch in November 2022. Older workers in those same roles stayed stable or kept growing. The authors call these young workers “canaries in the coal mine”: the first to feel the effects.

But read the caveats carefully, because the Stanford brief is scrupulous about them. The Federal Reserve began aggressively hiking interest rates in March 2022 — eight months before ChatGPT existed — and two papers find AI-exposed hiring began declining after that policy shift, not after the chatbot. Remote work also eroded the value of hiring juniors who learn fastest in person. When Brynjolfsson’s team added controls for these factors, the entry-level declines didn’t become notable until 2024 — by which point AI adoption and model capabilities had genuinely advanced.

So the honest read: hiring of young workers in AI-exposed occupations clearly fell around 2022, but AI can’t take all the credit. It’s the rare claim in this debate where even the skeptics concede something is happening — the question is how much of it is AI and how much is macroeconomics.

Where Jobs Are Actually Disappearing

The most granular picture comes from Bloomberry’s analysis of nearly 180 million global job postings from January 2023 to October 2025 — a dataset that got 200 points on Hacker News. Overall postings fell 8% in 2025, so any title that fell faster than that is losing ground to something specific. The losers cluster in one place: creative execution roles.

• Computer graphic artists: −33% (after −12% in 2024)

• Writers: −28% (copywriters, copy editors, technical writers)

• Photographers: −28%

• Journalists and reporters: −22%

• PR specialists: −21%

• Medical scribes: −20% — AI documentation tools are the obvious suspect

Notice the pattern: it’s the output-producing roles falling, while creative directors, creative managers, and other strategy roles hold up. The work that involves client judgment and complex decisions is resistant; the work that involves producing the artifact itself is not.

Here’s the twist: the steepest declines in the dataset have nothing to do with AI. Corporate compliance specialists fell 29%, sustainability specialists 28% — and chief compliance officers fell 37%. Regulation-driven roles collapsed faster than AI-exposed ones, because the regulatory environment shifted, not because a model got better at compliance. When a whole job market falls 8%, you have to separate the AI signal from the broader downturn. Even the AI-suspect declines are slower than they look: scribes fell just 2% in 2024 before this year’s 20% drop, so the jury is still out.

The Roles That Are Exploding

Flip the Bloomberry data around and the growth side is unambiguous. Machine learning engineer postings surged 40% in 2025 — on top of a 78% jump in 2024 — making it the single fastest-growing job title in the dataset. The whole AI infrastructure stack is hiring: robotics engineers +11%, applied/research scientists +11%, data center engineers +9%.

Indeed’s Hiring Lab tracks the same phenomenon at the posting level. Its AI Tracker — the share of US postings mentioning AI-related keywords — hit a record 4.2% in December 2025, while postings mentioning AI climbed 134% above February 2020 levels — against total postings that finished 2025 just 6% above that baseline. In some fields the shift is stark: nearly 45% of data & analytics postings now mention AI, versus about 15% in marketing and 9% in HR.

Demand is also skewing senior. Indeed found that 71% of the growth in US software development postings between May 2025 and May 2026 came from senior roles, and postings with AI in the title have surged to about 8% of all listings. Bloomberry saw the same shape: senior leadership demand is far stronger than middle management — the layer most exposed to automation.

And there’s a cautionary note for companies doing the “AI layoff” shuffle: Forrester research reported in October 2025 that half of firms that cut staff for AI planned to rehire — often at lower salaries. The jobs don’t vanish; they get cheaper.

The AI-Washing Problem: Layoffs Needing a Cover Story

The layoff data deserves its own skeptical section, because AI is increasingly the excuse. Challenger, Gray & Christmas — the firm that tracks every announced job cut — reported 153,074 cuts in October 2025, up 175% year over year, with year-to-date cuts above 1 million. Technology led the private sector with 141,159 cuts for the year. But Challenger’s own framing is careful: cost-cutting, softening demand, and pandemic-era over-hiring are all in the mix. Warehousing’s 47,878 cuts in October — a 48x jump from September — look far more like automation and overcapacity than like ChatGPT.

Fortune reported in January 2026 that AI layoffs increasingly look like “corporate fiction” masking a darker reality, and a May 2026 piece documented Wix, Block, Snap, and Atlassian citing AI for layoffs — a pattern one MIT professor says functions as a “cover story.” An independent analysis titled “Companies are lying about AI layoffs” pulled the numbers apart and found the same gap between the press release and the payroll data.

The official statistics back the skepticism. Only 5% of firms in Census Bureau surveys report any employment impact from AI — with equal numbers reporting gains and losses — and 80% of executives told the Atlanta Fed that AI investments haven’t changed headcount or productivity. A large Danish study linking worker-level and firm-level data found AI adoption restructuring tasks and time — but not employment, hours, or earnings. When the executives doing the layoffs say AI hasn’t changed their headcount math, believe them: the layoffs are about something else.

Productivity: The Missing Payoff

If jobs aren’t vanishing, what about the productivity miracle we were promised? The evidence is genuinely mixed — and the paradox is the most interesting part of this story.

In controlled studies, AI helps the workers who need it most. A large call center experiment found a generative AI assistant raised overall productivity 15%, with novice workers improving 30% — and no gain for top performers. GitHub Copilot studies found task completion 56% faster, again concentrated among less-experienced programmers. This is the “leveling” effect: AI compresses the gap between novices and experts.

But real-world measurement keeps complicating the picture. METR’s study of experienced open-source developers found participants were about 19% slower with AI — while believing they were 20% faster. And Glean’s survey of 6,000 workers found the new invisible job: “botsitting,” averaging 6.4 hours a week — feeding context to AI, checking outputs, cleaning up mistakes. 87% of workers use AI at work and 75% say it makes them more productive, yet only 13% say their organization performs significantly better because of it. Individual gains are being eaten by coordination costs.

That’s why aggregate productivity has been slower in the first three years of the AI era than during the 1990s IT boom — the same lag Robert Solow flagged in 1987 when he quipped that you could “see the computer age everywhere but the productivity statistics.” The technology arrives before the reorganization that makes it pay off.

What This Means for Your Career

Put it all together and the picture is neither apocalypse nor status quo. It’s a reallocation: the total number of jobs is roughly fine, but the composition is shifting underneath you.

LinkedIn’s own projection is the cleanest summary: the skills needed for the average job have changed 25% in the last several years, and LinkedIn expects that to reach 70% by 2030. As Lawit put it: “Even if you’re not changing jobs, your job’s changing on you.”

The workers feeling this most are the ones with the least leverage: new graduates competing for the junior roles AI does best, and workers in output-producing roles (writing, design, documentation) where models have genuinely gotten good. The workers gaining are ML engineers, AI infrastructure builders, and senior operators who know how to direct the tools.

One honest caveat before you calibrate your career on any of this: the studies cover roughly 2022 through 2025, and the HN comment section on the Stanford brief hammered on this point. Coding agents only started working really well in late 2025. The data we have is the era of chatbots assisting humans; the era of agents doing the work is only now beginning. The next round of studies may look very different — that’s exactly what the “normal technology” camp and the “world-altering by 2027” camp are arguing about.

What You Should Do About It

If you’re a worker, the data suggests a specific playbook rather than a panic:

• Stop competing with AI on output. Writing, design, and documentation volume is exactly where postings are falling 20-30%. Compete on judgment: client context, cross-functional decisions, the work AI can’t verify for itself.

• Get the seniority premium while it lasts. Demand is skewing senior across every dataset we looked at. The fastest way to protect your career is to move up the judgment curve — or position yourself as the person who directs the models.

• Learn the AI-adjacent stack. ML engineering, applied AI roles, and AI infrastructure are the only categories with +40% growth. You don’t need a PhD — the applied layer is where the demand is.

• If you’re a new grad, know the odds. Entry-level is the squeeze point, and it’s partly AI. Differentiate with demonstrated judgment and real project evidence, not coursework.

• Watch the agent transition, not the chatbot stats. Every number in this article describes the 2022-2025 era. The coding-agent wave that started in late 2025 is the variable that could make the next Stanford brief look very different.

The Bottom Line

The data-driven answer to “what is happening to jobs” is more boring — and more useful — than either side of the debate wants to admit. Aggregate employment is not collapsing. AI-exposed workers are not being fired faster than anyone else. But new graduates are getting squeezed, creative output roles are shrinking fast, and every remaining job is being rewritten — LinkedIn projects 70% of job skills will change by 2030. Meanwhile, the companies claiming AI caused their layoffs are mostly telling a convenient story, and the productivity gains that would justify the whole experiment are still stuck in the “botsitting” phase.

History says technological transitions take a decade or more to show up in the statistics, and the people who were loudest about the apocalypse have spent 2026 walking it back. But the tools that would change the math — agents that actually do the work, not just assist it — arrived right as the studies were being written.

If the data says there’s no AI jobs apocalypse so far, how confident are we that we’re not just measuring the last five minutes before one?

References

• Stanford SIEPR Policy Brief: “What is really happening to jobs? Separating AI hype from reality” (Mahoney, McEntarfer, Wahal, July 2026)

• Hacker News discussion of the SIEPR brief (300+ points, 377 comments)

• The Guardian: “The AI jobs apocalypse probably isn’t coming anytime soon” (Eduardo Porter, July 2026)

• Fortune: “Sam Altman and Dario Amodei are both walking back AI jobs apocalypse predictions” (May 2026)

• WSJ: “Big Tech Has Suddenly Flipped on the AI Jobs Wipeout Scenario” (July 2026)

• Apollo (Torsten Slok): “Where Is the AI Jobs Crisis?” (June 2026)

• FRED: Computer systems design and related services employment (BLS CES series CES5552000001)

• TechCrunch: “LinkedIn data shows AI isn’t to blame for hiring decline… yet” (April 2026)

• Indeed Hiring Lab: “January 2026 US Labor Market Update: Jobs Mentioning AI Are Growing Amid Broader Hiring Weakness”

• Bloomberry (Henley Wing Chiu): “I analyzed 180M jobs to see what jobs AI is actually replacing today” (Nov 2025, updated June 2026)