Five Doors, Five Bets

The four-stage framework in the last chapter is real, and worth doing. It also quietly assumed something that turned out not to be true: that there's one door to get through.

There isn't. Five of the largest technology companies alive are, right now, placing five different — and in places openly incompatible — bets on who gets to stand between you and your customer. Knowing which bet you're implicitly making, just by picking a vendor or ignoring the question, matters more than any feed cleanup.

Door one: OpenAI, the gateway that retreated

ChatGPT processes tens of millions of shopping queries a day, and OpenAI's original ambition was bigger than discovery — Instant Checkout aimed to complete the entire purchase inside the chat window. It didn't survive contact with reality. Flagship launch partner Walmart found conversion inside the chatbot running roughly three times below Walmart's own site, and OpenAI retired Instant Checkout about five months after launch. ChatGPT now does what it's genuinely good at — search, compare, recommend — and hands the shopper back to the merchant's own checkout. The door is open. It's a narrower door than originally advertised.

Door two: Google, the standard-bearer

Google took the opposite lesson from the same period: rather than owning checkout itself, build the plumbing and let merchants keep control. The Universal Commerce Protocol, launched with Shopify at NRF in January 2026, is backed by a genuinely unusual coalition — Target, Walmart, Etsy, Wayfair, Visa, Mastercard, American Express, Microsoft, Meta, and Salesforce all signed on. Merchants publish one capability profile; any UCP-aware agent can transact with them without a custom integration. This is the most "open door" of the group — literally open-source, Apache-licensed — and the one Salesforce has already built natively into Agentforce Commerce.

Door three: Microsoft and Perplexity, the fast followers

Less discussed, growing fast. Microsoft's Copilot Checkout has crossed 500,000 participating merchants. Perplexity runs its own in-chat purchase flow, Instant Buy, through a PayPal partnership. Neither has the volume of ChatGPT or Google's Shopping Graph yet, but both are gateway plays in the same mold — get chosen by someone else's assistant — and both matter more every quarter.

Door four: Amazon, the house that isn't open

This is the one that breaks the pattern entirely. While everyone else races toward shared, open standards, Amazon has built something deliberately closed. Its shopping assistant — Rufus, folded into a broader Alexa for Shopping in May 2026 — sits behind a wall: Amazon has not adopted MCP, UCP, ACP, or AP2 for it, and since November 2025 has been blocking dozens of outside AI crawlers from reading its own catalog.

Then there's the sharper edge: Buy for Me, which reaches outward — sourcing over 100 million products from more than 400,000 external merchants — and completes the purchase inside Amazon's own flow. If your product ends up there, Amazon owns the checkout, the payment, and the customer relationship. You become a fulfillment backend to a storefront that isn't yours, whether or not that's the relationship you'd have chosen.

WORTH KNOWING

For any retailer with meaningful Amazon exposure already, this isn't a fifth option to weigh evenly against the others — it's a live, ongoing fact about how much of your customer relationship you've already ceded, agent or no agent.

Door five: Anthropic, the one that isn't a gateway at all

And then, on September 2, 2026, Anthropic did something none of the above were doing: it open-sourced a free blueprint — Claude Commerce Agents — for building your own shopping agent, embedded in your own site, answerable to nobody's ranking algorithm but your own product decisions. No gateway to be chosen by. No feed to submit into someone else's black box. Shopify, Priceline, Visa, Mastercard, and Accenture are already building on it.

This is a genuinely different kind of door: not one you walk through, but one you build yourself.

The question this actually leaves you with

Four of these five doors ask some version of the same thing: format your data our way, submit it to our system, hope our undisclosed process picks you. The fifth asks something completely different: do you want to own the room, instead of hoping to be invited into someone else's?

Neither answer is obviously correct for every business — a small brand with no engineering budget has a very different calculus than a $2B retailer with an app already installed on ten million phones. But you can't make that call without knowing the five bets exist and pull in different directions. Most of what's currently being sold to retail leaders pretends there's just one door. There are five, and at least one of them — Amazon's — isn't really a door you get invited through so much as a room you might already be standing in without having noticed.

REFERENCES