The AI Layoff Reversal: When CEOs Confuse a Prototype for a Workforce

In 2024, the buy-now-pay-later company Klarna told the world that its AI assistant was doing the work of seven hundred full-time customer service agents; a year later, Klarna began quietly rehiring humans.

The reversal Klarna no longer talks about

The Klarna case is the cleanest stress test of the AI-replaces-humans theory the industry has produced, and it failed in public. In February 2024, the Swedish payments company said its AI customer-service assistant was doing the work of about 700 full-time human agents. Within one month of the assistant's launch, Klarna reported that it was handling roughly 75 percent of all customer chats, totaling about 2.3 million conversations. The numbers were treated as a proof of concept for the post-headcount enterprise.

By 2025, the story had inverted. Klarna chief executive Sebastian Siemiatkowski told Bloomberg that the all-AI approach had degraded customer experience, and that the company would begin rebuilding its human support team through 2025 and into 2026. His own words, in the Bloomberg interview, were precise about what had gone wrong: "as cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality." On the strategic pivot, he was equally blunt: "really investing in the quality of the human support is the way of the future for us."

What Klarna had built was not a workforce. It was a high-volume routing layer that performed well on the easy 75 percent of chats and lost the company quality on the rest. The rehiring is the receipt.

What Aaron Levie called it

The pattern Klarna exposed has a name now, supplied by an executive who runs an enterprise software company that benefits from AI hype rather than suffers from it. On May 27, 2026, TechCrunch reported that Box chief executive Aaron Levie said many tech CEOs are suffering from what he called "AI psychosis," the belief that AI agents can replace skilled employees outright. Levie's framing, quoted verbatim in the piece, was diagnostic rather than rhetorical: "CEOs are uniquely prone to AI psychosis because they're sufficiently distant from the last mile of work that still has to happen to generate most value with AI."

The mechanism Levie described is mundane. A founder or chief executive watches a model produce a polished prototype in a demo, and concludes that the prototype is the work. It isn't. The work is the code review, the contract verification, the integration, the exception handling, the customer recovery when the model is wrong. Those are the steps the daily team performs and the executive does not see. When the executive then announces that a team can be replaced, the gap between the demo and the production system gets filled by the customers, who notice immediately.

That argument was carried forward by Techdirt on June 9, 2026, in an essay titled "CEOs who think AI replaces their employees are just bad CEOs," which framed the 2026 layoff wave as a leadership failure dressed up as a technology success. The Techdirt piece treats AI as the financial cover for cuts that were already on the spreadsheet, not as a measured response to a working system.

The counter-example that nobody quotes correctly

IBM is the company most often held up to refute the Klarna story, and on the surface the numbers support it. IBM said its internal AskHR agent automates about 94 percent of routine human-resources tasks, including issuing pay statements, and it replaced several hundred HR roles in the process. If the AI-replaces-humans pitch were going to work anywhere, it would be in a process-heavy back-office function like HR ticketing.

But the headline from IBM is not the 94 percent. It is the fact that IBM's overall headcount grew. Chief executive Arvind Krishna told The Wall Street Journal that the freed budget from HR automation was redirected to hiring programmers and sales staff. His framing, paraphrased in the Journal, was that "AI freed up resources, which we redirected into hiring for roles demanding creativity, critical thinking, and human interaction."

That is a different theory of AI than the one Klarna ran. IBM did not treat AskHR as a workforce substitute. It treated it as a budget transfer from a low-judgment function to a high-judgment one. The company that automated the most ended up hiring more people, because the executive understood that the saved dollars had somewhere more valuable to go. The Klarna reversal and the IBM growth are not contradictory data points. They are the same data point seen from opposite ends of the prototype-to-production gap.

What MIT actually measured

The most useful number in this story is not a layoff figure. It is the failure rate. In August 2025, an MIT initiative called Project NANDA published a report titled "The GenAI Divide: State of AI in Business 2025." Drawing on more than 300 enterprise AI deployments, 52 case studies, and 153 leadership surveys, the report found that about 95 percent of enterprise generative-AI pilots had failed to deliver any measurable return on investment. Only about 5 percent of enterprise generative-AI projects produced measurable financial value.

The researchers were specific about the cause, and the cause was not the model. They concluded the failures were almost never the underlying AI itself. They were data readiness, workflow integration, and the absence of a defined outcome before building. MIT also noted that enterprises have spent an estimated 30 to 40 billion US dollars on generative-AI projects since the boom began, with only that 5 percent producing measurable value.

Read against the Levie diagnosis, the MIT figure stops being surprising. If executives are confusing prototypes for production work, then 95 percent of pilots failing to clear an ROI bar is what that confusion looks like in aggregate.

The 2026 layoff math

While the pilots were failing, the layoffs were not. The technology layoff tracker layoffs.fyi recorded more than 113,000 tech-sector job cuts in 2026 by mid-May, across 179 companies. Independent trackers cited by TechTimes put the total closer to 142,000 by late May 2026. Roughly half of those tracked layoff events in 2026 explicitly cite artificial intelligence, automation, or machine learning as a contributing factor.

The named cuts are not abstract. Amazon eliminated at least 30,000 corporate and technology positions between October 2025 and mid-2026, about 10 percent of its corporate workforce. Meta announced 8,000 layoffs on April 17, 2026, with internal guidance pointing to a possible 20-percent headcount reduction across the year. Oracle began laying off at least 10,000 employees on April 1, 2026, with analyst estimates of as many as 30,000 cuts, roughly 20 percent of its global workforce, before restructuring ends.

Set those cuts against the spending. A separate analysis cited by tech press in 2026 estimated that profitable tech companies have committed to a combined 700 billion US dollars in AI infrastructure buildout this year. The industry is firing workers and signing data-center contracts in the same quarter, while 95 percent of its own pilots fail to clear a return bar.

The Klarna reversal is the part of the story that is already settled. The MIT data suggests the prototype-to-production gap is not a Klarna quirk. It is the default. Somewhere, in a quiet office at dusk, there is one empty chair. The question for the executives who emptied it is whether they understood what the chair was for.

Sources

• Techdirt: CEOs who think AI replaces their employees are just bad CEOs (June 9, 2026)
• TechCrunch: Aaron Levie on "AI psychosis" (May 27, 2026)
• MIT Project NANDA: The GenAI Divide, State of AI in Business 2025 (August 2025)
• layoffs.fyi tech layoff tracker
• Bloomberg interview with Klarna CEO Sebastian Siemiatkowski
• The Wall Street Journal on IBM's AskHR rollout