Klarna's semi-reversal and Amara's Law

In early 2024, Klarna became the poster child of using AI to eliminate jobs. They announced that their OpenAI-powered support bot was handling two-thirds of their support requests, doing the work of roughly 700 human agents. They also said they were going to pause hiring and net reduce headcount, as they prepared for an IPO.

Now, they have changed their tune. It's not a complete reversal, as some headlines have suggested, but in fact a more nuanced approach than just "AI > human." Bloomberg:

Klarna Group Plc's chief executive says his pursuit of cost-cutting in customer service, fueled by advancements in artificial intelligence, has gone too far....

Co-founder Siemiatkowski said the firm is piloting a new cohort of employees "in an Uber type of setup" where they can log in and work remotely, with a view to eventually replacing "the few thousand human agents" that Klarna currently outsources....

The pilot has started small, with two of the new breed of customer-service agents live now, but the ambition is to tap into candidates such as students or rural populations. "We also know there are tons of Klarna users that are very passionate about our company and would enjoy working for us," he added.

"From a brand perspective, a company perspective," Siemiatkowski said, "I just think it's so critical that you are clear to your customer that there will be always a human if you want."...

"As cost unfortunately seems to have been a too predominant evaluation factor when organizing [our support function], what you end up having is lower quality," he said. "Really investing in the quality of the human support is the way of the future for us."

This admission comes at a time when a recent IBM survey of 2,000 CEOs found that "only 25% of AI initiatives have delivered expected ROI over the last few years, and only 16% have scaled enterprise wide."

Now, there may be some factors particular to Klarna. Over on HackerNews, Gergely Orosz floated the theory that last year's announcement was meant to frame Klarna "as an 'AI-first tech company' when it goes public, and not a 'buy now pay later loan company' because AI companies have higher valuations even with the same revenue." They might've pushed that strategy past its limit over the last year and discovered terrible results in customer experience surveys.

In any case, the megatrend remains true, that LLMs allow computers to do work that humans previously did, which will only accelerate as models and products improve. The demand for knowledge workers will go down as a cheaper, better, faster substitute good – knowledge agents – emerges. This is a capitalistic imperative: companies will be forced – either by their boards, shareholders, or competitors – to adjust their workforce towards digital agents. Just being real with you.

The takeaway from Klarna's situation and the study above isn't that LLMs have no role to play in organizational efficiency. The takeaway is that effective enterprise adoption of LLMs should be done carefully and responsibly, and despite the hype, it will take years for this transformation to yield significant results.

Amara's Law says, "We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run." The data points above are examples of overestimating the effect of a technology in the short run. Let's not commit the second half of the fallacy by over-extrapolating them.