There Can Be Only One

THE NUMBER: $68 billion. That’s Amazon’s advertising business last year. Most people think of Amazon as a store with a delivery fleet attached. Look at the number again. A big part of Amazon is a toll booth on getting found: sellers pay to rise in the search results, to win the little “sponsored” tile, to be seen by a human scrolling a page. Now send in a shopper who doesn’t scroll, doesn’t see tiles and doesn’t care who paid for placement. It reads the catalog, compares, buys and leaves. That’s what an AI shopping agent is. Keep the number in your head, because it explains why Amazon slammed the door in twelve days, and why, for everyone who isn’t Amazon, the same number reads as an opportunity.

Two doors, opposite directions

Meta (NASDAQ: META) launched Muse on September 8. By September 20, Amazon (NASDAQ: AMZN) was showing Muse users a pop-up instead of an order confirmation: “continued access by an unauthorized AI agent violates Amazon’s Conditions of Use.” Amazon’s complaint was that Muse browsed without identifying itself and captured and stored customers’ login credentials to get the job done. Fair complaint. Also a convenient one.

Then, the very next day, Shopify (NASDAQ: SHOP) did the exact opposite. It wired Shop Pay’s agentic checkout straight into Muse, so Muse can buy from any store that runs on Shopify. Same agent, same week, two of the biggest names in online commerce making opposite calls.

The feed wants you to pick a winner. I think they’re both right, and the reason they’re both right is the most useful thing to understand about commerce right now.

Why Amazon is right to say no

Amazon owns two things. It owns the customer (the relationship, the habit, the search box you type into without thinking) and it owns the logistics (the warehouses, the vans, the promise that it’ll be on your porch tomorrow). For twenty-five years those two things have reinforced each other. The customer shows up because the logistics are great. The logistics get better because the customers keep showing up.

An agent threatens to split them apart. If Muse becomes the front door, the place you go to say “I need paper towels,” then Amazon becomes a supplier the agent picks from. Amazon keeps the warehouse and loses the storefront. That’s still a phenomenal business. FedEx is a phenomenal business. It’s just a worse business than the one Amazon has today, and nobody walks voluntarily from owning both to owning one.

And the storefront is where the $68 billion lives. The ad business only works if a human looks at a page. An agent doesn’t look at a page. It strips the ads out of the exact transactions the ads were built on. So of course Bezos slammed the door. If I owned the tollbooth, I’d slam it too.

Why everyone else should say yes

Now flip it. If you are literally anyone other than Amazon (Walmart (NASDAQ: WMT) most of all, but also Target, Best Buy, the regional chains, the direct-to-consumer brands) you should be building for agents, not walling them out.

Every retailer on earth lost to Amazon the first time. They lost on selection, on convenience, on search, and most of all on habit. Amazon became the place you go before you’ve decided what you want, and once you own the moment before the decision, you own the category. An agent breaks that habit. It doesn’t start at Amazon. It starts with the need and goes wherever the answer is best. For a retailer that spent two decades losing the first click, that’s the first genuine rematch since 1999.

Walmart has clearly done this math. It signed up for OpenAI’s Instant Checkout in ChatGPT last year. Etsy (NASDAQ: ETSY) was the very first merchant platform on it. Shopify just did it with Meta. The pattern is obvious: the people who aren’t collecting Amazon’s toll are racing to be the stores agents can actually shop.

Open the door, not the vault

There’s one condition, and it’s the whole deal: the customer has to come through with the purchase.

We have seen this movie, and it ends badly. In 2000, Toys “R” Us signed a ten-year deal to let Amazon run its online store. It got the traffic. It paid Amazon handsomely for the privilege. Amazon got the customer data, the relationship and a front-row seat to how toys sell online. By 2004 the two were suing each other, and by 2018 Toys “R” Us was being liquidated. Borders made nearly the same deal in 2001, handing Amazon its website. It took the company seven years to build its own site again, and three more to go bankrupt. Both got the sale. Neither kept the customer.

So opening up to agents cannot mean handing the next Amazon your customers on a platter. It has to mean letting the agent browse your catalog while you keep three things: your checkout (you’re the merchant of record, not the agent), the customer’s name and email, and your right to remarket to that customer next week. That’s why the Shopify deal is smart and the Muse-scraping-your-logins approach isn’t. Shop Pay keeps the small shop as the seller. The agent brings the buyer. The store keeps the relationship.

Any agent that won’t pass the customer through gets the Amazon treatment. That’s not hostility. That’s the lesson of Toys “R” Us.

The part the agent can’t optimize

Now the confession that keeps this honest.

I know I could save money by spreading my shopping around. I’ve watched the price comparisons. I’ve read the articles. And I go back to Amazon time and again, because the truck rolls up by 6 pm and drops the stuff at my door, and I trust it. Prime isn’t a discount program. It’s a promise, kept thousands of times, until I stopped thinking about it.

That’s the thing the agent evangelists skip. An agent can make comparison shopping essentially free. It can find me forty cents off paper towels in half a second. What it can’t do is make twenty packages from twenty senders, arriving on twenty different days with twenty different return policies, any less of a pain in the ass. Search costs are going to zero. Fulfillment costs aren’t. The doorstep is physical, and you can’t prompt your way to a warehouse.

So the question for any shopper isn’t “can the agent save me money?” It’s “how much can it save me relative to the hassle it creates?” For a lot of everyday buying, the honest answer is “not enough.” Which means the logistics kings (Amazon and Walmart) aren’t going anywhere. They just stop being the only way to get found.

Stuck in the middle with you

In 1980, Michael Porter wrote the sentence that business schools have been teaching ever since. Win on cost, or win on being different. The company that tries to do a bit of both ends up “stuck in the middle,” not cheap enough to beat the cheap guys and not special enough to beat the special ones. For forty years that was a slow death. You could drift in the middle for a decade on location, habit and the fact that nobody could be bothered to compare.

Agents delete the grace period. They are Porter’s squeeze, automated.

Take the undifferentiated middle: somebody selling Bounty paper towels on their own website. On price, the agent finds a cheaper identical roll in half a second. On convenience, the Prime truck beats them to the porch. There is no third thing. They are a search result, and search results lose.

That leaves two places to stand:

Own the doorstep. Be the truck. That’s Amazon and Walmart, and maybe two or three others with the balance sheet to build it. If you’re not one of them, don’t pretend to be.

Be a category of one. Sell the thing there’s nothing to compare against. No agent can price-shop you down if there’s no second listing to put you next to.

How big is too big

There’s a wrinkle in “category of one,” and it’s the size of the market.

A unique product selling on Amazon purely for the logistics is fine until it gets big enough for Amazon to notice. The Wall Street Journal reported in 2020 that Amazon employees used third-party seller data to decide which products to launch under its own labels. Regulators in Europe and the U.S. went after the practice, and Amazon says it has policies against it. But the lesson for a seller is the same either way: once your category is large enough to matter to Amazon, you’re competing with the landlord.

And Amazon won’t bother below a certain size. A company chasing hundreds of billions in revenue doesn’t build a house brand for hand-painted book edges. That’s the safe zone, and it’s exactly the Shopify layer: millions of small shops with products too specific, too crafted or too small to be worth copying. Those shops have spent a decade paying Facebook and Google to be found. The agent is the first technology that works for them by default, because it reads the whole catalog instead of the paid slice.

Customer acquisition cost gets cheaper for everyone who lets agents in. It gets cheapest of all for the stores Amazon’s search never showed anyone.

Where the weird stuff lives

This is where eBay (NASDAQ: EBAY) and Etsy get very interesting, and I think the market underrates both of them in an agent world.

They hold a massive, very specific inventory of things that aren’t on Amazon and never will be. The buyers there aren’t looking for Bounty. They’re looking for that one thing. I collect tiki mugs. I’ve bought them on eBay and I’ve bought them on Etsy. Nobody’s tiki mug collection came from a sponsored listing.

Or take Maisie Matilda. She’s a young artist who paints fore-edge paintings: scenes hidden on the edges of a book’s pages that appear only when you fan them. She does them on editions of The Lord of the Rings and sells them, along with prints, on Etsy. It’s a nearly lost craft, one artist, one pair of hands. There is no Amazon Basics fore-edge painting. There’s no price comparison. There’s her, and there’s not-her. That’s what a category of one looks like.

Now imagine an agent whose only job is hunting 1950s Dodge hula lamps. That’s enormously valuable, because collecting is hard work. The right item appears once every few months, listed badly, photographed worse, in a marketplace you weren’t looking at that night. Serious collectors already live on eBay saved searches for exactly this reason. An agent is a saved search that can read the photo, spot a reproduction and tell you the seller has three other things you’d want.

Automate the vigil, keep the verdict

Longtime readers will catch a tension here. Yesterday we said: automate the chore, never the choosing. The hunt is the human part. So isn’t an agent hunting hula lamps exactly what we told you not to do?

No, and the distinction is worth making carefully. In collecting, the chore isn’t the choosing. It’s the stakeout: checking forty listings a night for three years in case the one you want finally shows up. Nobody loves that part. Collectors put up with it. The part they love is the moment of recognition, the judgment call: is it real, is it right, is it worth it, does it belong in the collection? That stays with you.

The agent sits the stakeout. You make the call. Automate the vigil, keep the verdict. Same rule as yesterday, applied to the one domain where the vigil is the brutal part.

The catch for the marketplaces

Here’s where I’ll push on my own enthusiasm.

eBay and Etsy win in the agent world only if they become something more than a place where listings live. Their old moat was the network: buyers came because the sellers were there, and sellers came because the buyers were there. An agent searching eBay, Etsy, a thousand Shopify stores and the estate-sale sites all at once weakens that. From the agent’s point of view, eBay is just one feed among many.

What an agent can’t replicate is trust. Is this Rolex real? Is this seller who they say they are? What happens if the mug arrives in pieces? eBay already has an authenticity guarantee on watches, sneakers, handbags and trading cards. That program is eBay’s version of the Prime truck at 6 pm. It’s the thing that makes a buyer, or the buyer’s agent, choose to complete the purchase there instead of anywhere else.

So the marketplaces have the same choice as everyone else, just one level up. Be the trust layer (authentication, buyer protection, seller reputation) and the agents route deals to you. Be just a listings page and the agents route around you.

What to do about it

If you run a store, three moves, today:

Write your agent terms before an agent writes them for you. One page, three non-negotiables: your checkout, your customer record, your right to remarket. Shopify’s deal with Muse is the template. Anyone who won’t pass all three through gets walled off.

Make your catalog readable by a machine. Agents don’t see your hero banner. They read specs, stock, price, delivery date and return policy. If a bot can’t answer “is this the right one, is it in stock, when will it arrive” from your product page without guessing, you’re invisible to the fastest-growing shopper in the world. That’s your new SEO.

Run the stuck-in-the-middle test. Take your top ten products. For each, ask whether an agent could find the identical thing cheaper in half a second. Every “yes” needs a reason to exist, whether that’s faster delivery, a bundle, a service or a story, or it’s a search result waiting to lose. The ones with nothing to compare against are your category of one. Double down on those.

There can be only one

In Highlander, the immortals spend the whole movie cutting each other down until one is left, because there can be only one. Commerce doesn’t work quite like that, but agents push it closer than it’s ever been. In every category, the agent is going to surface one answer: the cheapest identical item, the fastest delivery, or the only one of its kind. Second-best-at-everything doesn’t get surfaced. It gets skipped.

The long tail just got a salesperson. The middle just got a comparison engine.

Be the truck or be the only one. Everything in between is a search result.

Amazon Blocks Meta Muse AI Agent After 12 Days (Tech Insider, Sep 22, 2026)

Amazon blocks Meta’s Muse AI assistant in new standoff over agentic shopping (GeekWire)

Meta partners with Shopify to bring Shop Pay checkout to Muse AI (The Paypers, Sep 22, 2026)

Amazon and Meta exchange agentic fire (Fortune, Sep 22, 2026)

Amazon’s annual ad revenue passes $68B (Marketing Dive)

Amazon bars Muse from its store (The Implicator, Sep 22, 2026)

Why Amazon blocked Meta’s Muse (The Neuron, Sep 22, 2026)

Michael Porter, Competitive Strategy (Free Press, 1980)

Amazon Scooped Up Data From Its Own Sellers to Launch Competing Products (Wall Street Journal, Apr. 23, 2020)

OpenAI Instant Checkout in ChatGPT with Etsy and Shopify merchants (OpenAI, Sep 2025) and Walmart’s ChatGPT partnership (Oct 2025)

This Fore-Edge Book Painter Is Keeping a Critically Endangered Craft Alive (My Modern Met)

eBay Authenticity Guarantee (eBay).