AI was supposed to destroy jobs. Where’s the carnage?

The prediction was stark: artificial intelligence advancements would wipe out jobs en masse. “Half” of all entry-level white collar jobs would vanish, Anthropic’s CEO, Dario Amodei, said in May 2025. A month later, OpenAI’s CEO, Sam Altman, went further, foreseeing the end of “certain job categories”. Companies began citing AI in their layoffs. Workers organized. And students reconsidered their future careers.

But a year later, the mass carnage hasn’t shown up.

Even as AI capabilities have rapidly advanced and AI companies have hurtled towards trillion-dollar stock market debuts, economic transformation hasn’t kept pace, similar to previous tech revolutions, economists say. As a result, CEOs are reframing and softening their stances, suggesting AI augments workers rather than replaces them.

Despite the lack of mass job devastation, a shift is still under way: AI is changing the nature of work, with employers increasingly expecting job seekers to have AI skills. And over the long term, AI could shift more jobs to freelance and contract work as companies figure out which skills they do need, some economists predict.

Data from a recent Stanford Institute for Economic Policy Research analysis shows that AI hasn’t yet caused major job displacement. Since 2022, the year ChatGPT launched, the unemployment rate for the 20% of workers most exposed to AI rose by 0.77 percentage points, less than the 0.85 percentage-point increase for the least-exposed workers, the report showed. AI could be a factor in recent graduates’ rising unemployment, which hit 5.6% compared with the national average of 4.2% earlier this year. But factors including remote work and the unwinding of pandemic-era overhiring likely also played roles, the report states.

“Employment trends in the occupations [where] we would expect to see the impacts first are largely stable,” said Erika McEntarfer, fellow at the Stanford Institute and co-author of the report. “It took decades for the computer revolution to fully transform labor markets in the workforce, and what we’re seeing right now looks a lot like that.”

But accurately measuring AI’s impact on employment is a challenge. Government statistics are dated by nature and don’t track the impact of specific technologies, while private industry figures, though more current, are less comprehensive. So even though economists generally agree that AI will have an impact, they struggle to predict how big and when.

Jobs are changing, not disappearing

For now, AI’s biggest impact is not on the number of jobs, but on the nature of them. It’s consolidating roles, discouraging new hiring for tasks that can be automated, and raising the bar for who gets in, leaving unemployment numbers largely untouched.

Hiring trends show that AI is becoming more important for employers. About 74% of employers consider AI skills a strong advantage or requirement, with 13% requiring them company-wide versus just in technical roles, according to ZipRecruiter’s latest employer survey. Half of the polled employers expect candidates to already be practical or advanced AI users on day one. And sometimes the requirements don’t show up directly as “AI” in job listings, but rather as rising expectations around speed, quality and self-sufficiency.

“The clearest trend line is a rising bar rather than a shrinking pool,” said Nicole Bachaud, a labor economist at ZipRecruiter. “The labor market challenge for workers is increasingly about skills-matching rather than pure job scarcity.”

Employers have simultaneously added and cut within the same functions – tech, customer support, and business management and operations – signaling that “employers are still figuring out the exact skill set needed for success”, Bachaud added.

Nicholas Bloom, an economics professor at Stanford University, refers to this as turbulence in the job market. AI is destroying some jobs and creating others that have to implement, sell, fix and develop AI systems, he said.

Robert Seamans, a professor at NYU Stern who helped co-develop one of the standard measures used to gauge an occupation’s exposure to AI, categorizes AI’s impact into three buckets: jobs made obsolete, jobs created and jobs changed. “The third bucket is by far the biggest,” he said. “AI is changing and will continue to change the way most of us work, much in the same way that computers and the internet have.”

For instance, at AI coding platform Bolt.new, a three-person analytics team built an agent that analyzes data across all their systems, saving them 12 to 13 hours of manual work a week, said the company’s CEO, Eric Simons. With the help of an AI agent, their output is that of a 30-to-40-person team, he added.

“What it’s actually changing is how much one person can get done, and that shows up years before it ever touches a jobs number,” said Simons. “Their jobs got harder and way more interesting because they spend their time deciding which questions are worth asking instead of grinding out the answers.”

Workers may become more disposable

Still, some jobs are expected to become more temporary or easily replaceable, according to Paul Osterman, professor emeritus at the Massachusetts Institute of Technology and author of the newly released book Disposable Workers. More employers will likely turn to contractors and freelancers, instead of hiring more employees, as they figure out the required mix of skills needed for the AI future. This means more workers will be left without a career ladder. About 35% of the US workforce is already considered easily replaceable, according to his research. AI will only exacerbate this, he said.

As a result, more workers are trying to negotiate AI use in their collective bargaining agreements, said Tim Newman, senior vice-president of labor programs at the non-profit TechEquity. AI changes the kind of work people do as well as job quality, he said.

“That’s definitely what we’re hearing from workers,” Newman said, referring to jobs changing and deteriorating in quality. “A lot of people are experiencing [that], rather than full-scale displacement.”

It’ll likely take years for the full effects of AI on the economy to show up, Stanford’s Bloom said.

“A lot of the things that slow adoption … are very hard to accelerate,” he said. “It’s hiring new people, changing systems, changing job titles.”

The political climate surrounding datacenters and AI safety could slow adoption further, he said. “I could see politicians in the US, post-midterm, taking a strongly anti-AI turn.”

But not all jobs are in the same boat, and neither are their options for adapting, said MIT’s Osterman.

For high-skilled workers, the solution “is increasing your skill levels and external network, so you have some power in the labor market”, he said. “At the lower end, we’re going to need public policy to help protect people.”