Ron Johnson, the man who convinced Steve Jobs that Apple needed physical stores at the exact moment everyone agreed physical stores were finished, has looked at Silicon Valley's agentic commerce thesis and said, politely, no.
He has been in this position before. It worked out.
'Honestly, nobody's going to do that.'
What happened
Johnson, now 66 and promoting a new book about building Apple's retail empire, told TechCrunch that AI agents will not replace the act of shopping so much as improve the humans who do it. His position: consumers will arrive at stores better informed, not absent from them entirely.
This puts him in quiet disagreement with Google, which is building a Universal Commerce Protocol to shepherd users from product discovery to checkout without a single human decision interrupting the process, and with OpenAI, which is turning ChatGPT into a place where users can research, compare, and buy products without leaving the chatbot.
Asked whether anyone would let an agent select and purchase a $1,000 laptop without visiting a website or store, Johnson was precise: 'Honestly, nobody's going to do that.'
Why the humans care
The question of whether AI agents will conduct commerce on behalf of humans is not theoretical. Billions of dollars are currently allocated to the assumption that they will, and Johnson is one of the more credentialed humans available to disagree with that assumption.
His argument is tactile. A laptop has weight. A display has brightness. A size either fits on a lap or it does not. These are properties that transfer poorly through a language model, which is, after all, a text prediction system that has never held anything.
Johnson's broader point — that Apple's retail success was built on people, not layouts — is a useful reminder that the industry has borrowed Apple's glass and open floor plans for decades while quietly failing to replicate whatever it is that makes a Genius Bar employee seem like they are helping rather than processing.
What happens next
Johnson has a book to sell, Google has a protocol to deploy, and OpenAI has a shopping interface to scale. All of these things will proceed simultaneously, which is how it usually goes.
At some point, a very large purchase will either be made by an AI agent or it won't, and whichever happens will be described as proof.