Apple's store builder doubts AI shopping. Online went 1.1% to 17.1%

Ron Johnson built Apple's stores, and he does not believe people will hand shopping to AI agents. Asked whether anyone would let an agent pick and buy a $1,000 laptop unseen, he was blunt. His broader claim is softer and more interesting: AI will improve online shopping, but he does not know that it will change which way we shop.

That second sentence is the one with a record attached, because he made the same call at the start of the last channel shift. Here is what happened to which way we shop over his Apple years and since, on Census Bureau numbers:

• 2000, the year Johnson joined Apple: online was 1.1% of US retail sales.
• Second quarter of 2026: 17.1%, against 16.3% a year earlier.
• Still in stores today: 82.9% of the register.

Both readings of that are true at once, and the piece only works if you hold them together. Stores did not die, and 82.9% of American retail still runs through them, which is the half Johnson has always been right about. The channel mix also multiplied 15.5 times in the same period, which is the half the prediction missed. A forecast can be correct about survival and wrong about proportion, and the second error is the expensive one for anyone allocating capital.

“Honestly, nobody's going to do that.”

— Ron Johnson, Quoted by TechCrunch

Ron Johnson, founder of Apple's retail business, on letting an agent buy a laptop

A point a year

The pace is worth stating plainly too. Online gained 0.8 percentage points in the past year. Whatever agentic commerce does, it has to move a number that currently shifts under a point a year, which makes both the hype and the dismissal easier to check than either side would like.

Context on the speaker belongs here, and TechCrunch supplies it. After Apple, Johnson ran J.C. Penney from 2011 and was removed in under two years as sales fell. He then founded Enjoy Technology, which went bankrupt in 2022 and sold its assets to Asurion. He calls himself an AI optimist and says there is no substitute for human intuition. Both statements can be true; neither settles the forecast.

Where the argument actually lands

For our readers the interesting layer is underneath. An agent that genuinely buys needs a payment rail it can operate without a human at the keyboard, which is where stablecoin settlement and machine-readable checkout stop being theoretical. We measured that ground yesterday, when one platform's Tron payments turned out to carry the larger tickets. And the same displacement is already visible one step earlier in the funnel, where an AI answer cuts the click rate from 15% to 8%. Discovery moved first. Purchase is the part still being argued about.

Informational material, not investment advice. The retail share figures are US Census Bureau data for the second quarter of 2026 and for 2000; the comparison is ours.