Blog
The 8,500 Workers IKEA Refused to Lay Off
Call centers have a reputation. Mostly deserved. A 2026 stack of offshore scripts, hold music, and an "agent" who has not read your message. The industry consensus is that this labor pool is fungible. The cheaper you can swap it for AI, the faster your margin recovers.
IKEA read the consensus, shrugged, and did something else.
In June, Salary.run published the numbers: IKEA took 8,500 call-center workers whose jobs AI was eating, retrained them as interior-design advisors and remote sales consultants, and routed the AI underneath them as a copilot. Revenue per worker went up. Customer satisfaction went up. The unit shipped €1.3 billion in additional sales.
The lesson is not that AI does not take jobs. The lesson is that the businesses that treat labor as a sunk cost get the labor-market equivalent of a quarter-end fire sale. The businesses that treat labor as a balance-sheet asset they are actively rebuilding get paid back twice.
Most firms will refuse to copy IKEA. We know this because the alternatives are visible right now, and they are not pretty.
The Doom Lobby Missed the Reorg
There is a particular kind of essay that has been appearing on op-ed pages since the spring. The Atlantic published two of them in the same week, both worth reading for the symmetry of their mistakes.
"Three Ways to Think About AI and Jobs" tries to be neutral and ends up recapitulating the consensus. "America Is Headed Toward the Infinite Workweek" is the louder one: the prediction that AI agents collapse the workweek into a 70-hour blur because every task becomes a follow-up task.
The New York Times ran a parallel argument on the same theme, complaining that "AI doomaxxing" is bad for the economy. The irony is thick enough to spread on toast.
The doom lobby and the doomsayer lobby are both arguing from the assumption that the labor market of 2024 is the labor market we are stuck with. Neither has read the IKEA case. Neither has read PwC's 2026 AI Jobs Barometer. Neither has read Benedict Evans's framework on predicting AI exposure. All three were free.
CNBC reported on June 23 that Oracle shed 21,000 roles over the past year, with AI directly cited in the rationale. That is real. It is also a sample size of one company executing one strategy: replace a cost line, declare victory, do not mention the second-order effects. Evans's framework, by contrast, predicts the much more common outcome: jobs get decomposed, not deleted. Tasks inside the job split. The automatable half collapses into a prompt. The judgment-call half gets more valuable. PwC's data catches the same shape: two labor markets pulling apart. One flattening. One accelerating.
The Atlantic's "infinite workweek" essay is half right. The workweek is being renegotiated. But the renegotiation is not what the essay predicts. It is not AI extending every task into a new task. It is AI compressing the boring parts of the workweek so the meaningful parts can expand. A designer who used to spend four hours a day on calls now spends one. They spend the other three on the design work they were hired to do. That is not a 70-hour week. That is a reallocated 40-hour week with higher output per hour.
The doom essayists miss this because they do not work at IKEA. They do not work at the mid-cap logistics firm that quietly retrained 200 warehouse leads this spring. They do not work at the regional bank that promoted 30 branch managers into AI-supervisory roles instead of replacing them. These reorgs are not on op-ed pages. They are on balance sheets.
What the Smart Money Is Actually Doing
The interesting move is not the Oracle layoff. Oracle is a public company executing the cheapest strategy its board understands. The interesting move is what mid-cap companies in retail, logistics, and professional services are doing quietly.
The pattern is consistent. They are not firing. They are re-tasking. They are taking a role whose task mix is 60% automatable, stripping the automatable tasks into an AI agent, and giving the human the remaining 40% of the work, which is the part that actually required a human in the first place. The job title changes. The compensation changes. The person stays, because they have institutional memory and a network the AI does not.
IKEA called these people "design advisors." Other companies call them "account leads" or "client partners" or, in one case I saw this month, "decision architects." The naming convention is arbitrary. The economics are not.
The math is not subtle. If a call-center agent cost the company $X and produced $Y in revenue, and AI can now produce $0.7Y at $0.2X, the obvious move is to fire everyone and pocket the difference. The IKEA move says: keep everyone, let AI produce $0.7Y at $0.2X, give the remaining $0.3Y worth of work to the human who used to be a call-center agent and is now a designer, and watch the total per-worker output exceed the original $Y.
The companies that copy this will compound. The companies that fire will look great for two quarters and then discover they have no institutional memory, no customer-trust reservoir, and no bench to hire from when the next platform shift hits.
The Choice That Is Now On the Table
Every executive team has, as of June 2026, a binary decision to make about its workforce. They can run the Oracle play: cut, declare victory, watch institutional memory drain. They can run the IKEA play: re-task, retrain, capture the upside that the doom essays do not believe exists.
The Oracle play is cheaper in the short run and dramatically more expensive in the long run. The IKEA play is more expensive in the short run and dramatically cheaper in the long run.
Most boards will run the Oracle play. The reason is not stupidity. The reason is that quarterly earnings calls reward the cheap move and punish the IKEA move, and the CEOs who run the IKEA move will not get the credit for it until 2028.
The smart money is moving anyway. The smart money always moves before the credit shows up. The €1.3 billion is not a thought experiment. It is a receipt. Read it.
---
Filed 2026-06-25. Sources: Salary.run, CNBC, Benedict Evans, PwC, The Atlantic (×2), New York Times.