Remember when the internet was going to kill physical retail? Not the 2020 version of that prediction, the original. Late nineties, everyone with a modem and a PowerPoint had a slide showing malls flatlining by 2010. Bookstores were doomed. Record stores were doomed. The idea of walking into a building to buy a computer was considered a joke waiting for a punchline.
In 2000, Ron Johnson joined Apple to build a retail business anyway. You know how that turned out.
Johnson is 66 now, and according to the reporting making the rounds, he doesn’t follow Silicon Valley’s latest news. He’s focused on his retail insights and his new book. That’s it. The guy who arguably built the most successful square footage in the history of commerce is not tracking the AI shopping agent discourse, and honestly, that detail is doing more damage to the current hype cycle than any skeptical thread I’ve read this quarter.
Why This Registers As A Review Problem
I test AI tools for a living. A meaningful slice of what lands in my inbox is some flavor of agentic commerce: a bot that browses for you, compares for you, negotiates for you, checks out for you. The pitch is always framed as inevitability. Shopping is a solved problem, we just haven’t shipped the solution yet.
And every time, the same gap shows up. These products are built by people who find shopping annoying, for a hypothetical user who also finds shopping annoying. Which is a real user! I’m one of them. I would happily let an agent handle printer toner forever.
But that’s a narrow slice of commerce, and the teams building these tools keep mistaking it for all of it. The stores Johnson built weren’t efficient. They were slow on purpose. You went in without a plan and left with a thing you touched first. No agent replicates that because the agent’s entire value proposition is removing the touching.
The Detachment Is The Signal
What gets me is the shape of Johnson’s disinterest. He’s not out here writing rebuttals. He’s not on a podcast tour explaining why AI won’t work. He’s just… doing retail thinking and writing a book, while an industry three exits away insists his whole domain is about to be abstracted into an API call.
People who’ve actually built durable things tend to have this quality. They’re boring about it. They have one or two ideas they’ve tested against reality for decades, and they’re not especially interested in the quarterly narrative. Compare that to the AI commerce space, where the story has been rewritten four times in eighteen months: first it was chatbots, then recommendation engines, then autonomous agents, now it’s whatever the current term is.
A guy who spent a career on the same problem isn’t impressed by a category that changes its mind every two quarters. That’s not stubbornness. That’s pattern recognition.
What I’d Actually Want Tested
If you’re building in this space, the Johnson contrast should reframe your evaluation criteria. A few things I keep looking for and rarely find:
- Does the agent handle a purchase where the user doesn’t know what they want? Nearly all of them collapse without a clear query.
- Does it ever recommend not buying? Trust in retail is built on the employee who says the cheaper model is fine. No agent I’ve used does this convincingly, because the incentive structure won’t allow it.
- What happens after the transaction? Returns, service, the thing that breaks in month four. Johnson’s stores put a service counter in the back for a reason.
- Is there a reason a human would choose this over the existing five-tap process on their phone? “Fewer taps” is not a product.
Most of what I test fails on point two alone. The agent is a salesperson with a quota it won’t disclose.
The Honest Read
I’m not arguing AI has no role in how people buy things. Search is going to change. Discovery is going to change. Some of these tools will be genuinely useful for the boring 40 percent of purchasing that nobody enjoys. That’s a solid business and I’d review those tools kindly.
What I’m pushing back on is the confidence. The people predicting retail’s end have a losing record going back thirty years, and the guy who beat them last time isn’t even bothering to check the scoreboard. He’s writing a book about what he learned instead.
If your AI shopping product can’t survive a skeptical 66-year-old who isn’t paying attention to you, it probably can’t survive a customer either. Build something that works when nobody’s hyping it. That’s the only test that’s ever mattered, and it’s the one most of this category hasn’t taken yet.
🕒 Published: