Apple's retail architect draws the agentic commerce line at a $2,000 laptop

A laptop purchase is not one event. Discovery and comparison happen on a screen, payment can happen anywhere, and the part where a buyer holds the machine and picks between two screen sizes needs the machine in the room. "AI will never be able to have you physically experience a product," Johnson said ^[7].

Google's Universal Commerce Protocol is described as a standard for carrying an agent from product discovery through to checkout ^[3]. Apple designed its stores as places to buy a Mac, try products, learn how to use them, and come back for help when something went wrong ^[8]. Three of those four jobs sit outside the span a checkout protocol covers ^[20].

What Johnson says shoppers will do with agents is narrow the field and then walk in. "They'll just become more informed shoppers when they come to the store," he said ^[6]. The basis for that is design history, and the interview does not include usage data. He also points at staffing: Apple store employees are not paid on commission, Johnson said, and the intent was to remove the pressure to sell so staff could work out what a customer actually needed ^[9]. "The secret sauce for Apple has always been its people, the people in the store, and how they treat the customer," Johnson said ^[10].

The counterweight is in his own record. Enjoy Technology, the company he founded after J.C. Penney, brought technology products and setup services directly to customers' homes, filed for bankruptcy in 2022, and sold substantially all of its assets to Asurion ^[13]. He was betting that the home visit could replace the store visit. He is not arguing against the technology: "I'm a real believer in AI. I'm an AI optimist," he said ^[14].

For a merchant deciding how much to spend wiring up agent checkout, two checks per SKU are more useful than a view of the whole catalogue. One is what share of returns in this line cite fit, feel or size. The other is whether the buyer needs help after payment before the thing is worth anything. Where both answers are low, for replenishment, a known cable, a repeat order, an agent can carry the purchase end to end and the integration pays for itself in completed orders. Where either is high, the agent's job is to shorten research and hand off, and the measures are conversion after the handoff and the return rate on agent-placed orders. Those measures tell you which of the two you are running. A chat-session count does not.

OpenAI's version of this puts research, comparison and, in some cases, the purchase itself inside the chatbot ^[4]. Johnson was asked directly whether anyone would let an agent choose and buy a laptop at that price without ever seeing a website or a store, and said: "Honestly, nobody's going to do that" ^[5].