Silicon Valley's current retail thesis is that AI assistants will intermediate shopping — that people will tell a model what they want and let it find, compare and buy, displacing search, ads and storefronts alike. One of the people most responsible for the modern retail experience is unpersuaded.
TechCrunch reports that the executive who built Apple's retail stores does not buy the industry's bet on AI shopping. His scepticism is worth taking seriously precisely because it comes from someone who has run stores rather than models.
The case against the thesis is not that recommendation is hard to automate — it is that recommendation is not where retail is won. Selling physical goods involves inventory, supply chains, returns, store labour, fit and feel, and the trust that comes from a person or a brand standing behind a product. Large language models are good at language; they are indifferent to a shirt that fits badly, a delivery that arrives late, or a checkout flow that loses the customer.
There is also a demand-side question. Shoppers have repeatedly shown limited appetite for giving up control of a purchase to an intermediary, particularly when the intermediary's incentives are opaque. The incumbent assistant platforms already sit on a governance question — who pays for placement, and does the user know? — that retail regulators have begun to find interesting.
None of that makes AI shopping a dead end. Agents are genuinely useful for narrowing choices, comparing specifications and handling repetitive reorders. What the scepticism punctures is the maximalist version: the claim that the store, the salesperson and the brand relationship get replaced. The likelier outcome is augmentation — AI folded into shopping the way it is being folded into most workflows, removing friction at the edges while the physical apparatus stays where it is.
Caveat: this is an argument, not a data point in a controlled experiment, and retail's AI experiments are still early.
