AI Shopping Agents Hit Delays, Limits as Retailers Hesitate

I asked an AI agent to buy me a pair of running shoes last week. It asked four follow-up questions, timed out once, and came back with a different brand in a color I didn't pick. Technically, task complete. Practically, I could have done it faster myself.

That's the gap between the pitch and the product. Meta, Google, OpenAI and a handful of smaller players have all rolled out agentic shopping tools over the past year, pitched as a way to hand off the tedious parts of buying things, flights, hotel rooms, concert tickets, to software that handles it while you do something else. The demos are smooth. Real use is not.

Part of the problem isn't the AI at all. It's the retailers on the other end, who have to decide whether to let these agents through their checkout systems. Some are building for it.

Others are blocking it outright, worried about losing control over pricing, data and the customer relationship itself. Amazon has gone further than most, taking legal action against at least one company building shopping agents, a sign of how unsettled this arrangement still is.