For a while there, it looked like the robots had won. In early 2024, Klarna’s CEO Sebastian Siemiatkowski swaggered onto the tech stage with a bold claim: a single OpenAI-powered chatbot had replaced 700 human customer-service agents. Headlines swooned. Venture capitalists cheered. The future, it seemed, was finally here—and it didn’t need a paycheck, benefits, or empathy.
Humans handle the messier tasks
Fast-forward to this winter, and the swagger has softened into a sheepish shrug. Speaking to reporters in Stockholm last month, Siemiatkowski admitted that a stubborn slice of shoppers simply refuses to negotiate with a machine. “If you want to stay customer-obsessed, you can’t rely entirely on AI,” he confessed. Klarna’s next-gen bot—due out early next year—will still handle the easy stuff (password resets, delivery pings, the digital equivalent of asking where the restroom is), but anything messier—say, a double-charged credit card or a bridesmaid dress that arrived in the wrong shade of lilac—will be routed to carbon-based life forms. In other words, the 850-agent “equivalent” the bot supposedly represents will keep plenty of human colleagues within shouting distance.
Customers are unhappy
The Swedish fintech is hardly alone in its strategic retreat. Across the Atlantic, Verizon has spent the autumn quietly re-staffing its call centers after an AI-only experiment left four in ten customers fuming at the sound of synthetic voices. “Empathy is probably the key thing that’s holding us from having AI agents talk to customers holistically right now,” Ivan Berg, the executive in charge of Verizon’s AI push, told Reuters. The company still uses algorithms to triage calls and pre-fill agent screens—think of it as a very eager intern—but the final handshake, apology, or upsell is once again delivered by a human who can pronounce “Poughkeepsie” without sounding like a GPS having a stroke.
Even Zendesk, the San Francisco-based help-desk platform that has been cheer-leading generative AI since the ChatGPT boom, is pumping the brakes. President Shashi Upadhyay says his clients now let bots answer between half and four-fifths of routine tickets—an impressive stat until you remember that the remaining fifth-to-half involves blood-pressure-spiking edge cases. “The idea that generative AI can do everything is oversold,” Upadhyay sighs. The company’s internal data shows that the most successful deployments are the ones where the bot is trained to punt to a human at the first whiff of confusion, rather than circling the conversational drain until the customer rage-quits.
The Jagged Frontier
What went wrong? Researchers call it the “jagged frontier.” Large language models can now breeze through graduate-level math and debug Python scripts faster than most CS TAs, yet they still struggle to understand that “last week” technically ended yesterday and that Berlin’s boroughs don’t always come with neat postal labels. Euro Beinat, head of AI at the Dutch investment group Prosus, recounts watching an internal bot stumble over a seemingly simple request: how often a Prosus-backed food-delivery service arrived late with sushi in Berlin a few weeks ago. The model hallucinated neighborhoods, mis-counted days, and, in one memorable instance, included orders from Stockholm because the customer had once typed “Berlin” in a chat. People thought AI was magic. It’s not magic. There’s a lot of knowledge that needs to be encoded.
That encoding turns out to be expensive, tedious, and—ironically—human-labor-intensive. Banks are discovering that their data lakes are so inconsistently formatted that AI tools “read patterns that don’t exist,” like a fortune-teller hallucinating stock tips in tea leaves. Executives have decided that paying human underwriters is cheaper than teaching a model to understand every comma in a 200-page disclosure.
Brutal feedback
Consumers, meanwhile, have stopped being polite and started getting honest. A 20,000-person survey released this month by Qualtrics found that nearly one in five people who interacted with an AI support channel reported “no benefit at all”—a failure rate four times higher than in any other AI application. The gripe isn’t just accuracy; it’s the sense that companies are using chatbots to save money rather than solve problems. The result is a silent exodus: thirty percent of unhappy customers now leave without filling out a survey, up nine points since 2020. They simply vanish, taking their lifetime value with them.
All of which explains why 2026 is shaping up to be the year of the human rebound. Retailers from Ralph Lauren to Shopify are quietly testing “AI-augmented” concierges who can draft an email, pull up a receipt, and then—crucially—hit “send” only after a real person gives the nod. Mastercard’s chief digital officer, Pablo Fourez, calls the approach “agentic commerce,” a clunky phrase that basically means letting bots do the paperwork while humans handle the feelings. Even Amazon, long the poster child for friction-free automation, has added a tiny “Request a human” link beneath its Rufus chatbot; click it and the bot apologizes—yes, the bot apologizes—before summoning a flesh-and-blood rep within two minutes.
Stay connected with us to explore endless opportunities at Praxis Business School!
Visit our website at https://praxis.ac.in/ to learn more about our programs, admissions, and campus life. For any queries, feel free to reach out to us at https://praxis.ac.in/contact-us.
Follow us for the latest updates, insights, and success stories.
We look forward to connecting with you!