💼 Labor & AI
55% of Companies That Cut Workers for AI Now Regret It. We Calculated the Bill: $1.1 Billion.
Goldman Sachs says AI will displace 15 million American jobs. In 2025, it displaced 55,000. More than half the companies that made those cuts wish they had not. We used Challenger layoff data, Forrester's regret survey, and SHRM turnover costs to calculate what premature AI displacement actually cost the US economy last year.
Thirty thousand workers walked out of American offices in 2025 because their employers believed artificial intelligence could do their jobs. By mid-2026, more than half those employers had concluded they were wrong.
Three independent sources, taken together, tell a story nobody in the AI productivity narrative wants to hear. Challenger, Gray and Christmas counted roughly 55,000 AI-cited US layoffs in 2025, representing 4.5 percent of all announced job cuts and approximately 0.03 percent of total American employment. Forrester found that 55 percent of employers who made AI-driven cuts now regret the decision. And the Society for Human Resource Management puts the total cost of replacing a professional employee at 50 to 200 percent of annual salary, once you count severance, vacancy costs, recruiting, onboarding, and the productivity crater that opens when institutional knowledge walks out the door.
We combined these datasets, and the answer is not pretty. It is, however, quantifiable.
The Arithmetic of Regret
Start with 55,000 AI-cited layoffs and apply Forrester's 55 percent regret rate: that yields 30,250 jobs cut prematurely, positions that the companies themselves now acknowledge should not have been eliminated.
| Input | Value | Source |
|---|---|---|
| AI-cited US layoffs, 2025 | ~55,000 | Challenger, Gray & Christmas |
| Employer regret rate | 55% | Forrester |
| Prematurely eliminated jobs | ~30,250 | Calculated |
| Median salary, displaced roles | $72,000 | BLS (office/admin + professional) |
| Turnover cost, conservative (50% of salary) | $36,000 | SHRM |
| Turnover cost, moderate (75% of salary) | $54,000 | SHRM |
| Turnover cost, full-cycle (100% of salary) | $72,000 | SHRM / Work Institute |
At the conservative end, 30,250 premature separations at $36,000 each produces $1.089 billion. At the moderate end, the figure rises to $1.634 billion. At the full-cycle replacement cost that most HR researchers consider realistic for professional roles, it reaches $2.178 billion. None of these figures include severance packages, which typically run one to six months of salary, or the litigation and compliance costs that accompany involuntary separations.
| Scenario | Per-Worker Cost | Total Cost (30,250 workers) |
|---|---|---|
| Conservative (50% salary) | $36,000 | $1.09 billion |
| Moderate (75% salary) | $54,000 | $1.63 billion |
| Full-cycle (100% salary) | $72,000 | $2.18 billion |
One to two billion dollars. Gone. Shredded by companies that cut real workers to chase an AI productivity story that, in more than half of cases, did not materialize.
What the Prediction Said vs. What Actually Happened
On June 25, 2026, Goldman Sachs economist Joseph Briggs raised the bank's AI displacement estimate from 6 to 7 percent of US jobs to over 9 percent, translating to roughly 15 million workers displaced over the next decade. His methodology shifted from counting the "stock" of unemployed workers to measuring the "flow" of workers out of existing jobs, which inflated the number substantially.
Fifteen million sounds catastrophic, but in the first full year of aggressive corporate AI adoption, the actual number was 55,000: 0.37 percent of the ten-year projection, concentrated overwhelmingly in customer service, content production, and entry-level administrative roles. To reach 15 million at the 2025 pace would take 273 years. Nobody projects that.
| Metric | Goldman Sachs Projection | 2025 Actual | Gap |
|---|---|---|---|
| Jobs displaced (annual implied) | 1,500,000/year | 55,000 | 27:1 |
| Share of US employment | 0.9%/year | 0.03% | 30:1 |
| Unemployment rate impact | Up to 1% | Statistically undetectable | N/A |
JP Morgan put it bluntly in a recent analysis: of 1.1 million total announced US job cuts in 2025, AI was cited in less than 5 percent. Layoffs were "far better explained by government cutbacks, corporate belt-tightening and softening demand than by AI replacement."
The Klarna Parable
No company illustrates the overcorrection more vividly than Klarna. Klarna's CEO publicly claimed that its AI chatbot could replace approximately 700 customer service agents and froze hiring. Within months, the story became a case study in AI productivity presentations across the corporate world. Then quality metrics dropped, customer satisfaction fell, and by 2025 the CEO acknowledged that the company had "gone too far," initiating a quiet rehiring campaign in hybrid human-AI roles.
Klarna was not alone: Australia's Commonwealth Bank underwent a similar reversal, and across industries, the pattern repeated. announce AI-driven workforce reductions, watch output quality degrade, rehire under new job titles at higher salaries to attract workers who now had better options.
Gartner quantified the pattern in February 2026: only 20 percent of customer-service leaders actually reduced headcount due to AI. Meanwhile, the other 80 percent kept staffing steady or grew volume. Gartner predicts that 50 percent of companies that did cut customer-service roles citing AI will rehire for similar functions by 2027, often under newly minted titles like "AI Experience Coordinator" or "Human-AI Collaboration Specialist."
Why Companies Got It Wrong
Three forces drove premature AI cuts, none of them rigorous cost-benefit analyses.
First came AI washing: companies announced AI-driven layoffs to signal innovation to investors, boards, and the market. An earnings call that mentions "AI-driven efficiency gains" and "headcount optimization" reliably produces a stock price bump. What actually deployed was secondary to the narrative value of deploying it. Stock prices rewarded the story. Quality metrics told a different one.
Second came task-job confusion. Goldman Sachs itself notes that two-thirds of US occupations have "some exposure" to AI, but only 6 to 7 percent of workers face full displacement. MIT research confirmed that for computer-vision tasks, AI was economically viable for only 23 percent of exposed wages. Companies confused partial task automation with complete role elimination, treating "this tool can draft emails" as equivalent to "we no longer need someone who drafts emails, answers phones, manages calendars, resolves escalations, and maintains client relationships."
Third, the denominator problem. A customer-service chatbot that handles 80 percent of routine inquiries sounds like it replaces 80 percent of agents. But the remaining 20 percent of inquiries, the complex, emotionally charged, and high-stakes interactions, require 100 percent of an agent's skill and judgment. Automating the easy 80 percent does not reduce headcount by 80 percent; it changes what the remaining agents do and what skills they need.
A Generational Scar
Displacement hit hardest at the bottom of the ladder. Goldman Sachs data shows that the unemployment rate for 20-to-30-year-olds in tech rose by nearly 3 percentage points since early 2024, more than four times the overall rate, because entry-level coding, administrative, and content roles were the first positions companies cut, precisely the jobs that serve as on-ramps to professional careers, and precisely the cohort with the fewest alternatives and the least savings to bridge an 18-month gap.
On July 27, 2026, the Wall Street Journal reported that companies are reversing course. Sarah Franklin, CEO of HR platform Lattice, described the dynamic plainly: companies stopped hiring entry-level employees thinking AI agents could pick up the slack, then realized that humans are necessary to work alongside AI. "Just because you have coding agents doesn't mean you're not hiring engineers," she told the Journal. Across Lattice's thousands of clients, many are back in hiring mode for junior positions.
But the damage to the cohort is already done. A 20-year-old frozen out of the job market for 18 months does not simply resume their career trajectory when companies change their minds, because research on prior recessions consistently shows that graduating into a weak labor market produces earnings penalties lasting 10 to 15 years, a scarring effect that compounds through missed promotions, weaker professional networks, and the psychological toll of forced idleness at the start of adult life. For the cohort caught in the freeze, this created a miniature recession effect concentrated in the sector that was supposed to benefit most from the technology.
Limitations
Our calculation relies on Forrester's 55 percent regret figure, which comes from employer surveys whose sample size and methodology are not fully public. "Regret" does not always mean "will rehire the same roles." Some companies may regret the quality loss but still prefer the cost savings. Challenger data captures only announced layoffs, not quiet hiring freezes, contract non-renewals, or positions eliminated through attrition, all of which likely inflate the true displacement figure well beyond 55,000 and make our $1.1 billion estimate a floor rather than a ceiling. SHRM turnover costs are industry averages that vary enormously by role, seniority, and geography. And comparing Goldman Sachs's 10-year projection to one year of actual data is deliberately asymmetric, though we believe it is illuminating rather than misleading.
Strongest Counterargument
The bull case for AI displacement is that 2025 represents the beginning of a logarithmic curve, not the steady state. AI capabilities improved dramatically between GPT-4 and the current generation of agentic systems, and the next two to three years will see far more capable autonomous agents deployed in production environments. By 2028, 55,000 may look quaint. Goldman Sachs's methodology may be right about the destination, even if the timing is compressed. Companies that cut prematurely may have been right about the direction, just wrong about the speed, and the ones that held on may face a more wrenching adjustment later.
That is a reasonable position, but it does not change the present-tense arithmetic: real companies fired real workers based on capabilities that did not yet exist at the quality level required to replace them, and the measurable cost of that error exceeds one billion dollars.
Bottom Line
The AI job-displacement story is not wrong. It is early. Goldman Sachs, McKinsey, the IMF, and the World Economic Forum are almost certainly correct that AI will reshape the labor market over the next decade. But the companies that tried to get ahead of the curve in 2025 burned $1.1 to $2.2 billion in turnover costs, damaged their service quality, scarred a generation of entry-level workers, and are now quietly rehiring under new titles at higher salaries. AI will change work; nobody serious disputes that. But firing people faster than the technology matures is expensive, and 55 percent of the companies that tried it already know it.
What You Can Do
If you are a CEO or CHRO considering AI-driven workforce reductions, run the Klarna test first: deploy the AI system alongside existing staff for 90 days and measure quality, not just throughput. If output quality holds, reduce headcount through attrition, not layoffs. If it drops, you have your answer. If you are an entry-level worker in a role exposed to AI, document the tasks AI cannot do: the judgment calls, the exception handling, the relationship management, the moments when a customer needs a human. Your job security lives in the 20 percent that chatbots cannot touch. If you are an investor evaluating a company that announced AI headcount reductions, ask one question on the next earnings call: what is your 90-day quality delta since the cuts? If leadership cannot answer with data, the cuts were narrative, not operational.