A few months ago, my fiancée and I splurged on a pair of ebikes. We live in an especially hilly area of Atlanta, which is an especially hilly city, and we’d both just gotten bonuses at work, making the nearly £1,900 per bike feel, for a moment, digestible.

When my fiancée’s bike arrived on time, mine, bought separately from a different retailer, was delayed, then delayed again. Finally, one Wednesday evening, I got a text from FedEx confirming that my bike had been delivered to our address and signed for. This seemed impossible, considering that when the text arrived, I was standing in my kitchen, bicycle-less.

I checked outside our apartment; my package wasn’t there. My order confirmation showed it had been signed for by someone with the mysterious initials “M.M.” I set out to do that the following day by calling FedEx's customer service line for help. What followed was a monthslong descent into the bowels of customer service hell, during which I spent hours in virtual, chatbot-governed waiting rooms—with FedEx, the bicycle company, my bank, my credit card company, and even my local police department—desperately trying to find a real, living person who would speak to me, let alone solve my £2,000 problem.

Whether it’s due to organically bad experiences, intentional sludge, or some combination of both, it’s clear that shoppers are not happy with the state of AI-led customer service. In a report published in May and featuring consumers from the US, UK, and Canada, 59 percent said they were frustrated with AI customer service agents. Meanwhile, 85 percent said they’d prefer to speak with a real person.

Though it’s afforded gobs of hype, agentic AI is still in its infancy; Meta's Toolformer, a model that was considered groundbreaking for its ability to call upon external tools when in doubt, is just three years old. In my situation, a better chatbot may have directed my calls to the right person, or, given the amount of shipping data FedEx has at its disposal, it might have even tracked down my bike.