AI won't buy your laptop, but it will pick which one you buy
Apple's store architect says nobody will let an AI agent buy their laptop. He's probably right. But the research step is already gone, and that's the part Sri Lankan sellers are losing today.
AI shopping agents are getting billions of dollars in bets right now, and the man who designed the Apple Store thinks the bet is aimed at the wrong thing. Ron Johnson, 66, who joined Apple in 2000 and built its retail operation with Steve Jobs, told TechCrunch that on letting an agent choose and buy a $1,000 or $2,000 laptop for you: "Honestly, nobody's going to do that."
I think he's right about the checkout and wrong about what that implies. The transaction was never the valuable step.
🧠 What he's actually claiming
Strip the retail-legend framing off and Johnson makes two separate claims that deserve separate treatment:
| Claim | My read |
|---|---|
| "AI will never be able to have you physically experience a product" | True, and boring. It's a statement about atoms. |
| AI won't "change which way we shop" | Unsupported. It changes where the decision gets made, which is the part that pays. |
| Shoppers "will just become more informed" before they arrive | This is the real prediction, and it's already happening. |
| "The secret sauce for Apple has always been its people" | True for Apple. Not transferable, and I'll come back to why. |
Notice that his own third claim undercuts his second. If customers arrive already decided, the store has been demoted from decision venue to pickup counter with good lighting. That is a change in how we shop, even if a human still taps the card.
Key takeaway: The question isn't whether an agent completes your purchase. It's whether your shortlist was assembled before you ever opened a browser tab.
🔌 The blocker Silicon Valley keeps skipping: the rails
Here's the thing neither side of this argument tends to say out loud. Agentic checkout doesn't fail in Sri Lanka because customers distrust it. It fails because there is nothing for the agent to talk to.
Google announced the Universal Commerce Protocol in January 2026 to standardise how agents transact. OpenAI put shopping into ChatGPT Search back in April 2025. Both assume a merchant that has:
- A machine-readable product feed with stable IDs
- Prices as text, not baked into a JPEG
- Live stock status
- A card-payment endpoint an agent can hit
- A returns policy expressed in something other than a Facebook comment
Now picture a large slice of actual Sri Lankan retail:
| Layer | Global assumption | Common local reality |
|---|---|---|
| Catalogue | Structured feed, per-SKU URLs | Facebook album, price in the caption |
| Price | Text field, currency-tagged | Image overlay, or "inbox for price" |
| Stock | API field | "Let me check with the shop" |
| Checkout | Card endpoint | WhatsApp thread, then cash on delivery |
| Identity | Merchant account | A phone number |
An agent cannot buy from a phone number. So the near-term local risk isn't agents replacing your shop. It's agents being structurally unable to see your shop, while a competitor with a proper feed gets recommended instead.
🔍 The step you are already losing
Split any purchase into two halves and the argument resolves cleanly:
| Step | Who does it in 2026 | Value at stake |
|---|---|---|
| Discover, compare, shortlist, sanity-check price | Increasingly a model | High. Whoever wins this sets the choice set. |
| Confirm, pay, take delivery | Still a human, mostly | Low. It's plumbing. |
Johnson defends the second column. Nobody is seriously attacking the second column. The money is in the first, and the first is a text retrieval problem, which means it's winnable by a small team with no retail footprint at all.
A concrete version for a Sri Lankan buyer: the laptop question here is never "which chip is faster." It's "what does this actually cost landed, after duty." That's a documentable answer, so it's exactly the sort of thing a model will happily answer from whichever page states it clearly. If you'd rather check the arithmetic yourself, we keep a laptop import tax calculator and a mobile phone import tax calculator with the rates cited.
If you sell anything online, the cheap, dull, high-yield work this quarter is making your pages legible to a machine: Product and Offer JSON-LD, prices in the DOM, one canonical URL per SKU. You can eyeball how a page presents itself to crawlers and share cards with our meta tag preview before you spend money on anything cleverer.
⚠️ Why I don't take "people are the moat" at face value
I want to be fair to Johnson and also honest about his record, because it's the strongest evidence against his own thesis.
- He took the J.C. Penney CEO job in 2011 and was out in under two years. His own summary: "I applied a startup mentality to what needed to be a turnaround transformation."
- He then founded Enjoy Technology, which was essentially the Apple Store experience delivered to your door — the purest possible bet that trained humans are the product.
- Enjoy filed for bankruptcy in 2022. Asurion picked up the assets.
So the man arguing that people are the durable advantage has already tested that idea twice outside Apple, and it did not survive contact with someone else's cost structure. That doesn't make him wrong. It does mean "hire and train great humans" is a strategy with a known, brutal unit-economics failure mode, and you should not adopt it because a famous person said it warmly.
He also calls himself "an AI optimist" and grants that AI "will improve the online shopping experience." The disagreement with Silicon Valley is narrower than the headline suggests.
His position, fairly stated: AI changes the inputs to a purchase, not the venue. My position: the inputs are the business, and the venue was always a cost centre.
💡 What this means for you
If you're building or selling from Sri Lanka, here's what I'd actually do with this:
- Stop planning for agentic checkout. Start planning for agentic reading. The first is years out here and depends on payment rails you don't control. The second is live now and depends only on your HTML.
- Make every number quotable. Price, warranty period, delivery window, duty treatment — as text, on a stable URL. A model can only cite what it can parse.
- If you're a builder, the gap is the translation layer. Somebody has to turn Facebook-album commerce into structured feeds. That's an unglamorous, locally-specific, genuinely useful product, and it needs no frontier model to build.
- Don't buy either narrative wholesale. "Agents will do all the shopping" is a fundraising line. "People are the secret sauce" is a memoir line. The testable middle is that research moves to models and fulfilment stays physical.
- For your own purchases, use the asymmetry. Let a model do the comparison grind, then verify the two or three claims that cost you money if they're wrong.
The Apple Store was never really a shop. It was the highest-converting piece of content Apple ever made. That's the lesson worth stealing, and content is exactly the thing an AI-mediated internet reads first.
AI-assisted draft, reviewed and approved by Induwara Ashinsana before publishing. Sources are linked inline; if something here is wrong, tell me and it gets corrected.