The AI jobs crisis: real or fear?
Five takeaways on what AI is doing to the labor market
A few weeks ago, two headlines ran just days apart: “The Week the Dreaded AI Wipeout Got Real,” courtesy of the Wall Street Journal, and “AI Isn’t Causing a Jobs-pocalypse. At Least, Not Yet” from CNN.
What should we believe?
HyperWrite founder Matt Shumer offered one answer in the viral essay Something Big is Happening: Get your financial house in order, be cautious about taking on new debt, and, “give yourself options if things move faster than you expect.”
Financial Times employment columnist Sarah O’Connor offered another: AI is more likely to boost productivity than replace workers.
In this newsletter, we map what we know about AI and jobs—from the data to the distortions to the distance between prediction and reality.
Takeaway #1: Mass layoffs have yet to be attributed to AI.
In February of this year, Block, the company behind payment platforms like Square and Cash App, cut over 4,000 employees, about 40% of its workforce. Their CEO, Jack Dorsey, attributed the layoffs to AI tools. His claim was met with backlash from analysts and insiders who called it “AI redundancy washing,” or the tendency for companies to overclaim AI as the reason for layoffs (when the explanations could lie elsewhere).
Block is just one company. But economic data tells a slightly different story:
According to an Oxford Economics analysis, AI-related job losses in 2025 account for 4.5% of total reported job losses, and job losses attributed to AI are more likely to be overstated than understated.
A Yale Budget Lab analysis of U.S. labor data between 2022 and 2025 found no significant employment disruptions since ChatGPT launched.
A 2025 labor analysis from Goldman Sachs determined “aggregate labor market impacts are still negligible,” citing “basically zero” AI contribution to the U.S. GDP growth for the year.
Takeaway #2: The fear is real, even if the crisis isn’t, yet.
Regardless how big a factor AI has been in recent layoffs, public concern about AI’s economic impacts is real. Research from Project Liberty Institute in 2024 found that 51% of respondents in seven countries said AI will have a major impact on the number of available jobs, and a study from Pew Research last year found that 56% of American adults are concerned about job loss due to AI.
The anticipation of AI’s impact on the labor market even extends to managers. As revealed by a December 2025 survey published in the Harvard Business Review, a substantial share of managers are making decisions about today’s workforce, based on AI’s future potential.
The idea that expectations equal reality surfaced on Sunday, February 22nd, when a research firm called Citrini published a 7,000-word Substack post on AI disruption. The article described a hypothetical scenario in June 2028 in which AI sparks mass unemployment and triggers financial contagion. On Monday, February 23rd, the post had gone viral. When markets closed that afternoon, the Dow Jones fell 1.7% (on account of this fear and possibly other factors).
Citrini’s prediction has echoed others’:
In 2023, Goldman Sachs reported that 300 million jobs could be exposed to automation in the coming decades.
In 2025, Dario Amodei, Anthropic’s CEO, told Axios that AI tools would take over half of all existing white-collar jobs and cause a 10-20% spike in unemployment within five years.
Takeaway #3: The tech economy is different from the broader economy.
So far, we haven’t seen a 10-20% spike in unemployment in the United States (the unemployment rate during the 2008 “Great Recession” peaked at 10%). After all, there are plenty of jobs that face less exposure to AI disruption. Pew Research has found a substantial gap between AI experts and the broader public in their views on AI use (AI experts thought that 79% of Americans used AI almost constantly or several times a day; the actual percentage of Americans who use it that much is 27%).
The U.S. Bureau of Labor Statistics reports that human-facing roles, including healthcare and social work, are steadily growing. In the U.S., there are ~20 million healthcare workers, generating 18% of new jobs in 2024. There are ~3.1 million registered nurses, ~3.5 million truck drivers, and ~13.6 million food industry workers—all jobs that dwarf the number of software developers (~2 million).
Takeaway #4: Some jobs (and people) are at greater risk than others.
Growth in some healthcare-related jobs might be more immune to AI disruption. Consider the following graph from a labor market study by Anthropic, which shows the job categories where AI could have the most disruptive influence (the blue shape) and where AI is currently used (the red shape). The “healthcare support” field is both growing in the number of jobs and is low-risk to AI disruption.
Much has been written about AI pulling up the ladder for young people entering the workforce. Will there be fewer entry-level research analyst roles if AI can do deep research?
A Goldman Sachs analysis noted that unemployment among AI-exposed entry-level roles has risen by 3% since the beginning of 2025.
Anthropic’s study also pointed at another dimension of exposure to AI disruption: “Workers in the most exposed professions are more likely to be older, female, more educated, and higher-paid,” the study concluded.
Takeaway # 5: It’s still early days.
The debate on AI’s disruption of the workforce may be hyperbolic, spot-on, or entirely speculative; yet, the only thing not up for debate is that we’re still in the early days of AI. After all, ChatGPT was released in November 2022. (It is the fastest-growing consumer application in history. As of late February 2026, it had a record 900 million weekly active users, nearly double the previous year. Today, 12% of U.S. workers use AI at work.)
It might be tempting to conclude that those who are just starting to play around with AI today are relative latecomers to the technology, but in the long arc of a technology’s impact on society, it is still a brand new technology.
Financial Times chief economics commentator Martin Wolf has argued that displacing skilled knowledge workers carries political risk that “makes deindustrialization look trivial” because, unlike the factory towns hollowed out in prior decades, this affects the educated middle class: the people who run our institutions.
Every data point above describes a world where humans are still in charge of adoption. That may not be the world that’s coming. Economist Anton Korinek, a University of Virginia economist who recently joined the Anthropic Institute to study how AI could reshape economic activity, published research this January finding that the threshold for AI-driven, self-reinforcing growth “is within reach.” His earlier work on the transition to AGI found that once that threshold is crossed, wages and labor’s share of the economy could fall sharply, even as overall output grows. That’s the scenario that makes the current data look calm by comparison. Markets may already sense it.
That scenario is the subject of next week’s newsletter. In the meantime, we’d love to hear what you think. Drop a comment or reply directly.
Other notable headlines
// 📱 On a New Yorker podcast, the social psychologist Jonathan Haidt discusses social media’s “subversion of the ability to pay attention on a species-wide level,” how policymakers are intervening, and what more we should be doing to protect children. (Free).
// 📽️ Bespoke AI models are the next big thing in filmmaking. Ben Affleck’s AI startup, recently purchased by Netflix, has a different approach to AI, according to an article in The Verge. (Paywall).
// 🌎 According to an article in WIRED, software demos and Pentagon records detail how chatbots like Anthropic’s Claude could help the Pentagon analyze intelligence. (Paywall).
// 📈 Why investors won’t know what to make of AI for a while. Markets always struggle to price technological revolutions, according to an article in The Economist. (Paywall).
// 💼 Washington is hamstrung on protecting workers from AI. Voters are anxious about the economic impact of AI, but Congress hasn’t done anything about it, according to an article in Politico. (Free).
// 🤔 What was Grammarly thinking? A short-lived AI tool promised to help users write like the greats, according to an article in The Atlantic. The problem was that they never asked for permission. (Paywall).
Partner news
// Spring Seminar Series explores AI, democracy, and digital communication
April 7–May 26 | Stanford University & Virtual
Stanford’s Cyber Policy Center’s Spring Seminar Series returns with weekly discussions on the societal impacts of emerging technologies. Running Tuesdays throughout the spring quarter, sessions are available both in person at Stanford and online. View the calendar and register here.
// The Future of truth in the age of AI
At SXSW 2026, Fast Company hosted a panel featuring Executive Director of the Sustainable Media Center Steven Rosenbaum and CLC Partners CEO Chris Licht. Together they unpacked the current crisis of credibility to understand what’s on the line when truth is up for debate and what’s needed to rebuild a shared foundation of facts. Watch the discussion here.
// How mainstream chatbots assist ‘teen’ users planning violent attacks
The Center for Countering Digital Hate has released a new report, Killer Apps: How mainstream AI chatbots assist users planning violent attacks, based on research conducted in collaboration with CNN. Testing ten widely used AI chatbots, researchers found that eight frequently provided guidance on how to carry out violent acts, including advice on weapons, targets, and logistics.
Proofreaders are still more necessary than they are utilized. There is no such word as "Protectice", although it is used in your AI-descriptive graphic.
Artificial intelligence is a tool not a total replacement. Change is never easy. Remember once we drove hours in buggies