Layoffs: AI and the Math That Doesn't Add Up
On April 2, 2026, at 7:30 a.m. Chicago time, as on every first Thursday of the month, the phones of HR directors across half the world began to vibrate. The Challenger, Gray & Christmas report was released on time: 60,620 job cuts announced in the United States in March alone, a 25% increase compared to 48,307 in February. The top reason cited for the month? Artificial intelligence, responsible for 15,341 cuts—25% of the monthly total.
The number itself doesn't tell the whole story. Those who have followed these reports for years know this: Challenger measures layoff announcements, not executed layoffs. It’s a thermometer of managerial intention, not of the real market. But for that very reason, it is valuable: it anticipates rather than records. And what it has been anticipating for a few years now has a recurring name in corporate press releases.
The data for the first quarter of 2026 shows 217,362 announced cuts overall, the lowest total for a first quarter since 2022. It almost sounds like good news, until you remember where we’re coming from.
A Lasting Cycle, Not an Isolated Blow
2025 closed with 1,206,374 layoff announcements in the United States, according to final data from Challenger, Gray & Christmas. To find a worse year, you have to go back to 2020, the year of Covid, lockdowns, and mass forced closures. The comparison is enlightening: while in 2020 the cause was brutally external and visible, in 2025 the cause is more opaque, distributed, and narratively complex.
The January 2026 report had already sounded an alarm bell: 108,435 cuts announced in a single month, the highest monthly total since October 2025 and the highest January figure since 2009, when the world was still trying to figure out what had happened with the financial crisis. In that context, Andy Challenger, the company's head of external relations, observed that such large-scale cutting plans scheduled at the start of the year mean that decisions had already been made at the end of 2025, signaling a certain lack of optimism about 2026.
The pattern that emerges, looking at sectors, tells of a broader resettlement than the word "crisis" alone can contain. In 2025, the three hardest-hit sectors had been the federal government, retail, and technology. In 2026, already in the first quarter, the podium is occupied by technology (52,050 cuts), transportation (32,241), and healthcare (23,520). Each year a new composition, but the underlying direction remains unchanged.
The American Market Holds (For Now)
There is, however, a paradox worth naming precisely, otherwise there is a risk of confusing the thermometer with the disease. The U.S. Bureau of Labor Statistics published March employment data on April 3, 2026: unemployment rate stable at 4.3%, with 7.2 million unemployed, and 178,000 new jobs created in the non-farm sector. Healthcare added 76,000 jobs, construction 26,000, and transportation and logistics 21,000.
How does this data reconcile with the 60,000 layoffs announced in the same month? The answer is that we are looking at two different time scales and two different measures. Challenger measures the cutting plans publicly communicated by companies, often distributed over months or years. The BLS captures real employment at the time of the survey, including new hires. A company can announce 10,000 layoffs on Monday and hire 12,000 people over the next three months: both things are true, and both must be looked at.
Which doesn't mean all is well. The number of long-term unemployed—those without work for at least 27 weeks—increased by 322,000 over the past year. And federal employment, after its October 2024 peak, has lost 355,000 jobs, 11.8% of the total. The American market holds, but with pockets of distress that aggregate rates fail to show.
Tech in the Spotlight, but It’s Not Just Tech
The technology sector is the most visible protagonist of this season of cuts. In the first quarter of 2026 alone, it announced 52,050 layoffs—40% more than the same period the previous year and the highest quarterly figure since 2023, when the sector underwent one of its most severe purges following the hiring excesses of the pandemic biennium. Dell, Oracle, Meta with the Reality Labs division: different names, common logic—shifting resources toward artificial intelligence by taking them from other functions.
The layoffs.fyi tracker, which collects and geolocates cuts in global tech companies, offers a real-time mapping of this process. It is a tool born in 2022, when the sector began its first major post-pandemic correction, and has become one of the most cited references for those closely following the digital economy. The granularity of its data, updated almost daily, helps distinguish individual cases from structural trends.
But it would be a mistake to read this phenomenon as exclusively techno-centric. Challenger's report for the first quarter of 2026 shows that transportation recorded a 703% increase compared to the same period in 2025, with 32,241 cuts driven by airlines and shippers pressured by geopolitical volatility. Healthcare hit its historical record for a first quarter: 23,520 cuts, surpassing the previous record from 2023. The financial sector contributed 9,397 cuts, manufacturing joined in with waves linked to the automotive industry, and even the pharmaceutical sector announced 6,378 cuts already in the first quarter. The picture is of an economy reorganizing itself on multiple fronts simultaneously, not of a single sector in trouble.
AI as a Cause, or as an Excuse?
The most uncomfortable question of this phase is whether artificial intelligence is truly causing layoffs or whether it is becoming the preferred narrative to justify restructurings with more traditional roots: excess personnel hired during the boom, higher interest rates, correction of valuation multiples, slowing demand.
Challenger's data invites caution. In March, AI was the number one stated cause of monthly cuts, with 15,341 announcements out of 60,620—25% of the total. But on an annual basis, in the first three months of 2026, AI represents the fifth most common reason by volume, 27,645 cuts out of 217,362, about 13%. The primary reason remains overall market and economic conditions (45,103 cuts), followed by restructuring (37,916) and unit closures (37,405).
Since Challenger began tracking this data in 2023, AI has been cited in a total of 99,470 layoff announcements—3.5% of all plans communicated during the period. It’s a growing number—it was 3% as recently as February 2026—but it doesn't justify the alarmist narrative that often accompanies reports on the subject. As Andy Challenger summarized when commenting on the January report: it’s hard to understand how much impact AI is truly having on specific layoffs, because leaders talk about it and markets seem to reward those who cite it, but many cuts announced under that header have more prosaic origins.
This doesn't mean the impact is zero. It means it should be measured with greater honesty than corporate press releases, often written with an eye on analysts, tend to do.
The Math That Doesn't Add Up: When AI Costs More
Here comes the most interesting short circuit of the entire story: while companies use AI as a justification to cut jobs, a growing body of research suggests that directly replacing human workers with artificial intelligence systems can be much less cost-effective than the dominant narrative suggests.
A MIT CSAIL study showed that for vision-based tasks, only about 23% of wages linked to "exposed" jobs were economically advantageous to automate with current model costs. The point isn't just about output quality, but about the cost structure. In language models, every input token and every output token is accounted for. An AI agent that must complete a complex task—reading documents, generating hypotheses, calling tools, correcting errors, repeating steps—is not fixed-cost software: it is a continuous variable cost that grows with the complexity and ambiguity of the tasks.
The TheAgentCompany benchmark, built to evaluate AI agents on realistic activities in a business simulation, found that the best model tested managed to independently complete 30.3% of tasks, with a score of 39.3% including partial results. In other words: in more than half the cases, the agent fails, and someone must still intervene to check, correct, and take responsibility for the result. It’s not replacement: it’s a new category of work being added—that of the automation supervisor.
The most counter-intuitive data comes from a randomized METR study on experienced developers working on open source repositories familiar to them: those with access to AI tools took 19% more time than those working without them, despite expecting to save time. The time gained in code generation was more than offset by the time spent checking, reviewing, and correcting outputs. It’s the same paradox that anyone who has used these tools intensively in professional contexts knows well: AI accelerates the easy part and slows down the hard part, which is often the part that counts.
Gartner estimated that over 40% of agentic AI projects will be canceled by the end of 2027 due to rising costs and unclear business value. The problem is not asking in the abstract whether AI costs less than an employee, but calculating the real cost to complete an activity with acceptable quality, including tokens, supervision, errors, training, system integration, and prompt maintenance. This calculation, when done honestly, significantly diminishes the theoretical economic advantage.
Focus on Italy: The ISTAT Snapshot
Italy looks at this scenario from a position that deserves its own careful reading, without automatically transplanting the American narrative. The ISTAT press release on the fourth quarter 2025 labor market offers a picture that goes in the opposite direction of alarmism.
At the end of 2025, there were 24,121,000 employed people in Italy, an increase of 37,000 units in the fourth quarter alone compared to the third. Year-on-year, employment growth was 185,000 units, +0.8%. The unemployment rate stood at 5.5%, down 0.5 points from the previous year; the annual figure is 6.1%, also declining. The number of unemployed decreased by 138,000 units over the year, nearly 8.9% less. The most dynamic component was that of permanent employees, which grew by 1% year-on-year, and self-employed workers, which grew by 3%.
The picture is not without shadows: temporary employees fell by 8.6% during the year, signaling a market segmentation that continues to weigh, and the number of inactive people between ages 15 and 64 grew slightly again. But the overall balance is that of a market that has not suffered the shocks recorded in the United States.
This doesn't mean AI isn't touching Italian labor. It means the timing and transmission modes of the phenomenon are different, linked to a productive structure with a fabric of small and medium-sized enterprises less exposed to the mass layoff announcements that dominate American headlines, and to a regulatory framework that makes collective layoffs a long and regulated process, involving union consultation, preventive procedures, and precise deadlines.
Invisible Flows: Hires, Transformations, Terminations
Employment and unemployment rates are snapshots. To understand the real health of a labor market, one must look at flows: how many people enter, how many leave, with what type of contract, and in which sector. The Ministry of Labor's Mandatory Communications—the data that companies must transmit for every change in the employment relationship—are the most granular source available in Italy for this type of analysis.
What emerges from the available quarterly reports is a market in continuous shuffling, where terminations do not necessarily coincide with layoffs in the technical-legal sense. In Italy, the distinction between individual dismissal, multiple dismissal, and collective dismissal is not just formal: it determines procedures, protections, corporate costs, and timing. A collective dismissal—one affecting five or more workers within one hundred and twenty days—requires a procedure with trade union organizations and, in many cases, examination by regional labor authorities. This explains why private American trackers, built for a market where companies can announce 10,000 redundancies in a morning, are not suitable tools for reading the Italian labor market.
The practical consequence is that the impact of AI on Italian employment will likely manifest more gradually, less visibly in headlines, and more distributed over time: not so much in large waves of announced layoffs, but in the failure to replace those who leave, in the transformation of sought-after profiles, and in the growth of required skills for roles that were once accessible without specific technical training.
Two Speeds, Many Open Questions
There is a scene in the finale of the second season of Severance, the Apple series that transformed the metaphor of work-life division into dystopian architecture, where the characters discover that the separation they thought they were suffering was also, in part, something they had chosen. The suggestion is useful: the resettlement of the labor market we are experiencing is not just something happening to us. It is also the result of collective choices about how to invest, which technologies to adopt, and which risks to accept.
The data we have looked at tell two parallel stories that should not be confused. The American one is the story of a market in accelerated transformation, with waves of cuts that in 2025 exceeded 1.2 million announcements, driven by a mix of genuine restructuring, past excesses needing correction, and a growing but still minority share of real technological replacement. The Italian one is the story of a slower, more regulated market, with 185,000 new employees in 2025 and a falling unemployment rate, which will absorb the waves of transformation in its own time and forms.
The questions that remain open, however, are the same on both sides of the Atlantic. Does AI truly create more jobs than it destroys, as its evangelists claim? Are the estimates from Goldman Sachs, McKinsey, and the like regarding the imminent automation of hundreds of millions of jobs proving accurate or too aggressive? And above all: if MIT CSAIL data tells us that only 23% of "exposed" tasks are already cost-effective to automate today, and if the TheAgentCompany benchmark tells us that the best available agent independently completes only 30% of realistic tasks, then how well-founded is the certainty with which many companies are building their restructuring plans on the hypothesis of an AI capable of replacing entire functions?
The point is not to reassure; the transformation is real and its impact will be significant. The point is not to confuse AI marketing—which promises total autonomy to sell platforms and convince markets—with the operational reality of systems that still fail in more than half of complex tasks. As in every major technological change, who wins in the first phase is not necessarily the one who is right in the long term, but the one who manages to build the most convincing narrative in the short term.
The final bill, for workers and companies, is yet to be settled.