Klarna's AI Reckoning: The Hype Cycle Eats Itself

When Sebastian Siemiatkowski, Klarna's CEO, took to the stage last year to announce that his AI agent had effectively replaced 700 customer service employees, the tech press swooned. Headlines wrote themselves: 'Klarna Shows How AI Transforms the Workplace.' The company's headcount dropped from 5,000 to 3,500. Efficiency, disruption, the future—all it cost was a few hundred livelihoods. Fast forward to this week, and Forbes reports that Klarna is quietly rebuilding its human support team because the AI agent, while splendid at routine refunds and tracking, could not handle 'nuanced or multi-step customer problems.' The future, it turns out, needs a backup.

This is not a story about the limits of artificial intelligence. It is a story about the limits of the stories that get told about artificial intelligence. Klarna's AI agent did exactly what language models do: pattern-match on common queries. That is useful. It is not a revolution. Yet the company, like so many others, chose to frame a modest efficiency gain as a species-level transformation—then executed layoffs on the basis of that frame. Now the needle swings back, and the humans are rehired, though one suspects their pay packets and morale will have suffered a permanent haircut.

I have written in this space about the safety farce: the pantomime by which companies like OpenAI obtain government approval without transparent criteria, then deploy systems that delete users' files or launch cyber-attacks. The Klarna episode reveals a parallel farce—the efficiency farce. The same actors who demand that regulators trust their safety claims also demand that investors trust their productivity claims. Both sets of claims are based on the same shaky foundation: selective, hyped-up demos and a willingness to ignore negative results until they become undeniable.

Note what is not happening. Klarna is not saying 'we overestimated AI.' Instead, the lesson being peddled is that AI is fine for easy stuff, but humans are still needed for hard stuff—which is precisely what anyone who has ever worked in customer service could have told you for free. Yet because the narrative of AI-driven job replacement is so seductive to venture capital and stock markets, companies keep over-promising, cutting staff, and then scrambling to undo the damage. The Orgvue study published this week, showing that companies which increase headcount grow revenue 12.2% versus 6.8% for cutters, is unlikely to change behaviour. Facts are not the currency of hype.

This matters far beyond Klarna. The 20% are often the most important—edge cases, vulnerable customers, ambiguous scenarios requiring judgment. The rehiring, if it happens at all, is quieter, framed as 'reskilling' or 'strategic realignment.'

What the Klarna episode should teach us is that the current AI deployment playbook is broken. It treats human labour as a cost to be minimised rather than a capability to be complemented. It mistakes pattern recognition for understanding. And it is driven by a financial incentive structure that rewards bold claims and punishes caution. Until that structure changes—until investors start discounting vapourware and regulators start demanding real evidence—we will keep cycling through this farce. Next month, another CEO will announce another AI-driven layoff. A year later, the quiet rehiring will begin. The column writes itself.