Over the past two years, a specific type of corporate email has gained popularity. Phrases like “strategic pivot,” “operational efficiency,” and, these days, almost always something about artificial intelligence are included. It usually arrives on a Tuesday morning at around nine in the morning. Three weeks ago, a friend who has worked for her company for seven years, received positive reviews, and is in good standing received one of those emails.
Customer service position, Fortune 500 company. She wasn’t fired, according to the subject line. According to the statement, the business was “evolving its workforce model to align with AI-driven operations.” She is still unsure of the precise meaning of that. It turns out that many of the businesses are not sending these messages either.
| Topic | AI-attributed tech layoffs and return on investment in artificial intelligence |
| Total tech layoffs (2025) | Over 180,000 workers cut across the industry — averaging 489 jobs per day |
| AI ROI failure rate | 95% of companies investing in AI report zero measurable return on investment (MIT, 2025) |
| Total AI investment (est.) | Between $30 billion and $40 billion spent on AI initiatives with minimal profit impact |
| Salesforce cuts | 4,000 customer support jobs eliminated; company later confirmed hundreds were redeployed internally |
| Klarna reversal | CEO initially credited AI with replacing 700 agents; later stated “we have made 0 layoffs due to AI” |
| Microsoft layoffs | 15,000 workers cut in the past year; employees report pressure to adopt AI “whether we like it or not” |
| Amazon layoffs | 30,000 employees let go in six months; teams using AI tools that were “not fully functional” |
| Academic finding | Yale Budget Lab analysis of U.S. labor data (Nov 2022–Jul 2025): AI has not caused widespread job displacement |
| NY Federal Reserve | Only 1% of service firms reported AI-related layoffs in the past six months — down from 10% the previous year |
| Key experts cited | Fabian Stephany (Oxford), Ethan Mollick (Wharton), Stuart Russell (UC Berkeley), Stephan Rabanser (Princeton) |
The numbers of layoffs are truly astounding. In 2025 alone, more than 180,000 tech workers were laid off. Microsoft eliminated 15,000 jobs. In six months, Amazon lost 30,000 employees. In February, Block laid off over 4,000 employees, or about 40% of its workforce. Atlassian and Pinterest, two smaller players, eliminated 10% and 15% of their headcounts, respectively. AI is mentioned at some point in practically every press release.
The technology that is purportedly taking on the workload of thousands of departing employees is the new strategic priority and the cause of the restructuring. “The maximum hype you have right now, which is that AI is replacing people, is not true. However, the evidence for that claim is, to put it politely, thin.” However, it’s also untrue that AI won’t ever pose a threat to employment. It will be challenging. — Wharton School’s Ethan Mollick
Just 5% of AI pilots extract any quantifiable value, according to an MIT study that examined 300 public AI deployments and drew from interviews with 52 executives. 95% of businesses that invested between $30 billion and $40 billion in AI projects reported no appreciable increase in profits. Nothing. Somewhere between the boardroom slide deck and real operations, the tools malfunction, the workflows break, and the promises vanish.
The researchers found that businesses frequently invest in AI for sales and marketing, which are known for having poor returns, while ignoring back-office automation, which has much higher returns. This could be a reflection of real uncertainty about where AI truly operates. It might also be a reflection of something more pessimistic.
AI and labor researcher Fabian Stephany, an assistant professor at the Oxford Internet Institute, has kept a close eye on this trend. He comes to the direct conclusion that many businesses are using AI as a scapegoat for traditional cost-cutting measures.
He is supported by data from the New York Federal Reserve. Only 1% of service companies reported laying off employees as a result of AI in the previous six months, compared to 10% the previous year, according to the most recent survey. Minimal AI causation, record layoffs. The math just doesn’t work.
For a while, Klarna was the epitome of AI displacement. The CEO of the Swedish fintech company, Sebastian Siemiatkowski, spoke on podcasts and in interviews about AI taking over the tasks of 700 customer service representatives after the company laid off 40% of its employees. The story was clear and dramatic. Then, in what could be charitably described as an explanation, Siemiatkowski stated that the business had “0 layoffs due to AI.”
He clarified that the cuts were the outcome of natural attrition after a hiring freeze that started in 2023. A chatbot did not take the place of the employees. They simply weren’t replaced. It was interesting to watch that reversal unfold in real time, with a company that had come to represent the end of AI jobs subtly reversing the entire concept.
Another case that merits careful consideration is Salesforce. 4,000 jobs were eliminated as a result of CEO Marc Benioff’s audacious public declaration that AI could handle 50% of customer support tasks. Afterwards, a Salesforce representative admitted that hundreds of those workers had been “redeployed into other areas.” Translation: The business relocated employees. That doesn’t mean AI will take the place of people. That is a reorganization dressed in more futuristic attire.
After realizing that AI couldn’t handle the true complexity of human interactions, IBM and Klarna, at different times, completely changed their approaches to AI customer service. They spent millions, laid off employees, and then began hiring again.
The experience of AI is far less revolutionary on the factory floor, or in this case, the open-plan office, than the press releases portray. AI was producing code more quickly than humans could review it, according to a former engineering supervisor at Block. He claimed that with the same number of reviewers, there was three times as much code. “We were falling behind on reviews,” he stated. Because AI-generated code may appear authentic while subtly introducing conflicts or bugs, human review is important.
Without supervision, speed is a problem in and of itself. Similar experiences were reported by a senior designer who was recently let go from Amazon Web Services. His team had been experimenting with two internal AI tools, both of which were still in early testing and not yet fully operational. “None of this is ready yet,” he thought to himself as the cuts appeared. How will all of this work be completed?
These businesses also face a more subdued form of pressure that is more difficult to measure but is frequently discussed. When it came to the adoption of AI, former Microsoft employees spoke of a “feeling of being watched”—a general perception that adopting the tools wasn’t optional, even if they slowed things down. Similar tales were shared by Amazon employees, who felt subtly threatened that their jobs might be in jeopardy if they didn’t adopt.
According to Amazon, using AI is not required. However, the cultural signal seemed to be quite different. “You don’t want to be known as the person against AI,” stated a former employee of Microsoft. It’s difficult to ignore the dilemma that results from employees being forced to use ineffective tools in companies that have already started reducing headcount in the name of those same tools.
When you take a step back from the noise, the research actually demonstrates that we are in an experiment with no clear hypothesis or end point. After three years of examining U.S. labor data since ChatGPT’s launch, Yale University’s Budget Lab concluded that the AI disruption that has dominated headlines has just not materialized on a large scale. According to Stephan Rabanser, a postdoctoral researcher at Princeton who has studied AI reliability, consistency is still the main obstacle since existing AI tools still have trouble consistently producing the same right answer under various circumstances or with various users.
“Reliability will be a key limiting factor,” he stated. High-quality training data is getting harder to come by, according to Berkeley AI researcher Stuart Russell, and chatbots frequently give confident answers even when they don’t have the information needed to do so.
Whether this is an actual turning point in labor history or just a particularly aggressive round of cost-cutting dressed in the language of the future is still up for debate. In certain proportions, both may be true. AI is actually transforming the way technical work is done; Block reported comparable figures, and Google credited AI with writing half of its code. These are actual changes. However, there is a significant gap between “AI helps with coding” and “AI eliminated thousands of jobs.”
Businesses seem to be bridging that communication gap while navigating the more complex reality on the ground. For the majority of them, the payout is not coming on time. The cost of that discrepancy is borne by the workers who are caught between the story and the reality.