In February 2024, Klarna’s CEO Sebastian Siemiatkowski told the world that the company’s AI assistant had taken on the work of 700 customer service agents. Headcount fell from 5,500 to 3,800. The story became the most cited example of AI replacing humans at scale. Boards across Asia, Europe, and the US used it to justify their own restructuring conversations.
Eighteen months later, Klarna was quietly rehiring. By February 2026, Siemiatkowski publicly admitted that the company had gone too far. Customer satisfaction had cratered. Software engineers and marketers were being pulled onto support lines to plug the gaps. The CEO who had once claimed AI could do every job, including his own, was now telling Bloomberg that the quality of human support was the new priority.
Klarna is not the cautionary tale of a single company that bet wrong on AI. It is the cautionary tale of a thinking error that most AI-driven reorgs are making right now.
The data on the thinking error
A Harvard Business Review article published in January 2026, authored by Thomas Davenport and Laks Srinivasan, surveyed 1,006 global executives in late 2025. The numbers landed hard. Sixty percent of organisations had already reduced headcount in anticipation of AI. Only two percent of those organisations had reached the point where AI was actually doing the work the cut humans used to do. Fourteen percent had AI solutions ready to deploy. Eleven percent were using AI in production.
The math is uncomfortable. Six out of ten companies had cut. Two out of a hundred had a working AI replacement for what they cut. The other 58 were either betting the gap would close before customers noticed, or quietly absorbing the work back into the humans who remained.
Davenport and Srinivasan called this AI washing. Companies using AI as the narrative cover for financial restructuring that they would have done anyway. Recent research from agentic AI vendors confirms the pattern: 55 percent of companies that executed AI-driven layoffs now regret the decision. Gartner projects that 40 percent of agentic AI projects will be cancelled outright by 2027.
This is not a problem about AI capability. It is a problem about how leaders are framing the question they are trying to answer.
Also Read: Why Japan’s booming AI market is harder to crack than it looks
What work-first design looks like
The companies getting AI team design right are not the ones starting with the question “how do we restructure for AI?” They are starting with a different question. What does the work itself actually want to look like now?
I call this Work-First Design, and the difference shows up in the outcomes.
At Tripadvisor, AI agents now handle 90 percent of incoming customer queries autonomously. The headline number sounds like Klarna’s. The strategy underneath is the opposite. Tripadvisor did not set out to eliminate human roles. The company set out to free the human support team for strategic work that required judgment, creativity, and relationship-building. The 90 percent automation rate enabled a 100 percent reassignment of human attention to work AI could not do. Thumbtack and ClickUp built similar models.
McKinsey research from 2025 found that companies which fundamentally redesign their workflows around AI are three times more likely to capture real value from the technology, and they generate twice the AI usage per employee. The redesign companies are also the ones building pod structures that work. Meta’s Reality Labs reorganised a large group into AI-native pods with roles like AI Builder, AI Pod Lead, and AI Org Lead. Engineers were expected to operate with broader range. Pods were required to own outcomes rather than isolated tasks.
The pods are not the point. The work redesign underneath is the point. Putting “Pod Lead” titles on top of a workflow that has not been redesigned just renames the old problem in new vocabulary.
This is where most reorgs fail. The leaders running them have been sold a structure. Pods, agents, AI-native teams. The structures are real and many of them work. But they only work if the work has been redesigned to fit. Drop a pod structure on top of a customer service workflow that still requires emotional judgment on 30 percent of cases, and you get Klarna. Drop the same pod structure on top of a workflow where AI genuinely handles 90 percent and humans handle the judgment-heavy 10 percent, and you get Tripadvisor.
The structure looks identical from the outside. The outcomes are not.
The Klarna pattern is going to repeat
The reason Klarna is going to keep happening is that work redesign is harder, slower, and less narrative-friendly than structural reorg. A reorg announcement makes the board happy. A six-month work redesign with no headlines does not.
Leaders are also being pushed by the wrong signals. Compensation benchmarks now reward AI fluency at every level. The PwC Global AI Jobs Barometer reports a 56 percent wage premium for AI-skilled workers. The labour market is telling leaders to hire AI talent fast and restructure around them. The temptation is to do exactly that, then figure out the work design later.
Later is when the customer satisfaction scores collapse. Later is when the engineers get pulled onto the support phones. Later is when the CEO has to tell Bloomberg that the strategy was wrong.
Also Read: AI is changing global expansion, but it cannot standardise local markets
I built and exited a SaaS company without taking venture capital. That meant I never had the budget to throw structure at problems. Every team I built had to match the shape of the work, because there was no spare capital to absorb a wrong design. That discipline turned out to be the most valuable constraint of my operating years. The companies that are now learning this lesson under AI pressure are learning it the expensive way.
The good news is that the lesson is learnable. The bad news is that the leaders most likely to ignore it are the ones with the most capital to throw at the problem first.
Three questions for leaders rethinking team design in 2026
What is the actual shape of the work after AI is genuinely doing what it can do, and what is left for humans?
If you removed every “AI” job title from your reorg plan, would the structure still solve a real problem, or does it only make sense as an AI narrative?
If your customer satisfaction scores or your output quality drop 15 percent in the six months after the reorg, what is your specific plan to recover them?
If the answer to the third question is “we will rehire,” you are not redesigning. You are doing a Klarna in slow motion. The cost of that mistake has now been documented in detail. There is no excuse left to make it.
—
Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.
The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.
Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.
The post Why most AI driven reorgs are solving the wrong problem appeared first on e27.