Automate the Task, Not the Person

In late 2024, Klarna’s CEO told the market that AI could already do the work people do. The Swedish fintech paused hiring, let its workforce fall from roughly 5,500 to 3,400, and promoted a chatbot it said was handling the work of about 700 customer service agents.[1] Within months, service quality had slipped, complaints rose, and the company was asking engineers and marketers to help answer customer queries. By 2025, Sebastian Siemiatkowski admitted to Bloomberg the company had “gone too far,” and Klarna began hiring people back, this time promising customers a human would always be reachable.[1]

Klarna’s communications team now calls the reversal an evolution rather than a retreat. That framing is the tell.

The lesson underneath it is more reassuring than the headlines suggest. Don’t let AI make your headcount decisions. Let it inform them. AI can show you which tasks have changed. It cannot tell you where your people create the most value.

The tool was never the variable

The companies that got this wrong and the ones that got it right were not separated by their technology. A support chatbot is a support chatbot. What separated them was a decision about what a job actually is.

Ravin Jesuthasan, who sits on the World Economic Forum’s steering committee on work and has made this argument for over a decade, says the real question was never which jobs AI replaces. It is “what work will be redefined, and how.”[6] His method is to break a job into its tasks, automate the routine ones, and rebuild what remains around the people. A job is not one indivisible block you keep or delete. It is a bundle of tasks, a few of which a machine can now do and most of which it still cannot.

Heather McGowan frames the same idea from the human side. Think augmentation, not automation, because “the tool is only as good as the hand that uses it.”[7]

The data supports the caution. Gartner predicts that by 2027, half of the companies that attributed headcount cuts to AI will rehire for the same work under new titles.[5] More revealing still, in Gartner’s own survey only about one in five customer-service reductions were primarily driven by AI. Most were ordinary economics.[5] A lot of companies, in other words, are pinning a decision on AI that AI never made.

That leaves three paths: reverse, rebalance, and reskill.

What a rebalance looks like

Salesforce walks the path most companies actually occupy. On a podcast in late 2025, Marc Benioff said the company had taken its support organization from about 9,000 people to 5,000, “because I need less heads.”[2] AI agents now handle roughly half of support interactions, with customer satisfaction holding about steady.[2]

But the fuller picture is more careful than the soundbite. Salesforce redeployed hundreds of those people into sales, professional services, and customer success, and its total headcount sits at a record, above 83,000.[2] This was a reshape, not a shrink. Benioff himself later cautioned that blaming AI for tech layoffs was too convenient.[2] The same executive who said he needed fewer heads also warned against handing the machine the credit, or the blame, for the decision.

What a reskill looks like

IKEA took the path that turns automation into growth. Ingka Group, its largest franchisee, deployed a chatbot named Billie that came to resolve about 47% of customer enquiries, some 3.2 million interactions.[3] The obvious move was to cut the agents whose routine work the bot had absorbed. Ingka did the opposite. It studied what Billie could not handle, found the unresolved queries clustering around interior design, and reskilled 8,500 call centre workers into remote design advisors. That new service produced 1.3 billion euros in a single financial year, about 3.3% of sales, with a target of 10% by 2028.[3]

Two honest caveats keep this from being a fairy tale. The 1.3 billion is the entire remote selling channel, not solely the output of the reskilled group. And in 2026 IKEA did cut roughly 1,650 corporate roles, which it tied to soft sales and tariffs rather than AI, with the reskilled service workforce not in scope.[3] The reskilling decision stands on its own merits. It was never a promise that the company would never make a cut anywhere.

IBM shows the same principle in a different function. Its AskHR agent automated a large share of routine HR queries, and a widely shared story claimed the company had replaced 8,000 people and then hired them back. That story is wrong. The real number was a few hundred HR roles, and IBM’s total employment rose, because the freed investment went into engineering, sales, and roles its CEO described as critical-thinking work. “Our total employment has actually gone up,” Arvind Krishna told the Wall Street Journal.[4]

What this means if you are still close to the work

At a 40-person company, one wrong call lands harder than it would at a 40,000-person one. That is precisely why the founder has the advantage here. You are close enough to the work to see it as tasks rather than titles, and to move a capable person into the part of the job a machine cannot do.

The downside of getting it wrong is concrete. The work does not disappear when the role does. It resurfaces, often within months, and someone has to be there to handle it. In the gap, quality slips in front of customers, and your standing with the people you would most want to hire takes the hit. Gartner’s forecast that half of AI-attributed cuts will be reversed is not a prediction about robots. It is a prediction about decisions made too quickly.

McGowan’s line is the one to keep: “learning is the new pension.”[7] The durable move is to automate the routine and invest in the people doing everything else.