Something genuinely fascinating happened in the last week of September, and it happened so fast that if you blinked you missed the entire philosophical split that's going to define the next era of e-commerce.

Shopify opened its checkout to AI shopping agents. As in: an AI running inside someone's browser can now read the checkout page, update the delivery options, apply a discount code, and place the order. On your store. Without you doing anything to enable it. (Or, notably, anything to disable it.)

Amazon, approximately a few days earlier, told Meta's brand-new AI shopping agent Muse to get lost.

Same agent. Two doors. One opened, one slammed. If that doesn't tell you everything about how these platforms see the future, I don't know what will.

Shopify extended something called WebMCP into its checkout flow. MCP — Model Context Protocol — is essentially a structured menu that tells an AI agent exactly which actions it's allowed to take on a page. Instead of the agent squinting at your beautifully designed checkout like a confused tourist trying to read a menu in a foreign language, the page hands it a clean list: here's the delivery option, here's the discount field, here's the "place order" button. Call these functions. Done.

The agent can't add items to the basket, and it can't enter new card details. Payment still runs through Shop Pay or a saved method. Shopify's documentation is quite firm that the agent must show the shopper the total and get their explicit permission before completing the purchase. (Which is reassuring, because "surprise, I bought you a sofa" is not the kind of AI autonomy anyone asked for.)

This builds on a cautious first step earlier this year where agents could search storefronts and build carts but were stopped at the checkout door — like a very efficient personal shopper who's allowed to pick out your outfit but not actually hand over the credit card. Now the handoff runs all the way through payment.

Shopify has shared some internal test data claiming the protocol approach completes every checkout attempt while browser automation (where the agent just guesses its way around the page like the rest of us) missed a few, and did so faster at roughly half the cost. Those numbers come from Shopify, testing on Shopify's own shops, on a single AI model. Worth noting. Worth holding lightly.

The reason a routine checkout update matters is that it lands squarely in the middle of a question every major commerce platform is currently having a quiet existential crisis about: when an AI agent does the buying, do you want to be the one running the agent, or are you willing to be the store the agent shops at?

Everyone, obviously, wants to be the agent. The company whose agent does the shopping sits closest to the customer, sees the intent, captures the relationship. The company whose catalogue merely gets browsed risks becoming a vending machine that a robot walks up to. Nobody wants to be the vending machine.

Amazon has answered this with characteristic subtlety: they want to own the agent everywhere. They've deployed Rufus as their in-house shopping assistant, folded conversational commerce into Alexa, and built a feature called Buy for Me that sends Amazon's own AI agent to purchase products from other retailers' websites without the shopper ever leaving Amazon's app. Meanwhile, they've blocked outside agents from operating on Amazon itself, are actively litigating against Perplexity over whether a user's AI assistant can act on a site where that user holds an account, and — in the move that makes the whole picture click — told Meta to remove Amazon from Muse's experience entirely.

Read together, this isn't contradictory. It's actually impressively coherent: Amazon wants to mediate every purchase, whether it happens on Amazon or not. Their advantage was never the chatbot. It's the fusion of purchase history, browsing behaviour, review sentiment, pricing, inventory, and delivery logistics into a personalisation layer that no outside agent can touch. An external agent can scrape a product listing. It cannot see what you buy every six weeks. Keeping other agents out protects that. Letting Amazon's own agents roam the open web extends it.

(In other words: "We'll shop at your store, but you can't shop at ours." Which, as negotiating positions go, is certainly bold.)

Shopify's position is almost exactly the inverse, and it follows from where they sit rather than from any particular fondness for openness. Shopify doesn't own the end customer. Its customers are the merchants. It has no demand-side data moat to defend because the demand belongs to the shops running on it. What Shopify owns is the rail — the catalogue, the checkout, the payment flow through Shop Pay. So its interest is the reverse of Amazon's: make Shopify merchants the easiest inventory for any compliant agent to transact with, and be the layer every agent calls when it's time to pay.

Opening the checkout to browser agents serves that for agents in general. The deep deal with Meta's Muse serves it for one agent in particular — enabling agentic checkout through Shop Pay across all Shopify stores, processed at merchants' existing payment rates. Shopify's share price jumped roughly seven per cent on the news before giving some of it back, which tells you the market liked the thesis even if it hasn't quite decided how much to pay for it.

Amazon can afford to close because it holds the demand and the data. Shopify almost has to open because its asset is the network and the rail, and a rail earns nothing if traffic gets diverted around it. Both strategies are rational. Neither is obviously the clever one. They're each playing the hand they were dealt, just in opposite directions.

The fascinating part is what happens in the middle. Through Buy for Me, Amazon may send its own agent to transact on your Shopify store. Meanwhile, your listings on Amazon remain reachable only through Amazon's agent. So the customer relationship ends up sitting in a completely different place depending on which channel the sale runs through. That's not a traffic question. That's a control question.

For Shopify merchants: orders placed by AI agents may start appearing in your dashboard without you having opted into anything. That's the convenience and the catch in one neat package. Shopify's documentation doesn't yet clarify whether you can switch individual agent tools off or separate agent-placed orders from human ones in your reporting. A sale that a machine completes in seconds looks very different in the analytics from a person who browsed for ten minutes, and your conversion rates, session metrics, and attribution models were all built around the second kind of shopper. The practical task right now is to watch for these orders as they emerge and understand how they distort your numbers before drawing conclusions from them.

The deeper shift is about what an agent actually pays attention to. When a protocol lets an AI read structured product data and call a checkout function, your carefully designed product page matters less and the quality of your underlying data matters more. A machine selecting on a shopper's behalf weighs price, availability, delivery speed, returns policy, and clear specifications. It does not linger on your hero image. It does not respond to a well-judged piece of visual merchandising. (Sorry to everyone who spent six weeks perfecting their above-the-fold lifestyle photography. The robots don't care about your aesthetic.)

For Amazon sellers: while Amazon keeps outside agents out, the agent that matters on its platform is Amazon's own. Optimisation moves away from indexing for keywords and towards being the product the system recommends when it interprets a shopper's intent. Review sentiment, catalogue completeness, pricing discipline, fulfilment reliability — the trust infrastructure an AI weighs when deciding what to put forward.

You now have two buyers to satisfy: the human, and the machine shopping for the human. They want different things and read different signals. The human can be moved by a brand. The machine, for now, mostly cannot.

Making your products legible and trustworthy to agents — through accurate feeds, consistent pricing, honest inventory, and clear terms — is becoming a distinct discipline from making them appealing to people. And the two platforms that dominate online commerce have just given you opposite playbooks for how they think it should work.

The questions that will decide where the money lands are still wide open. If agents converge on shared protocols, does the value accrue to whoever owns discovery (the agent that chooses) or whoever owns the transaction rail (the checkout that completes the sale)? If your storefront becomes a set of structured functions any agent can call, what's left to compete on beyond price and reliability? And the governance questions — whether merchants can see or limit which agents transact on their behalf, whether a shopper's agent inherits the shopper's permissions, who answers for an order a machine placed — are entirely unsettled.

Shopify has decided it would rather be shopped by agents than bypassed by them. Amazon has decided the opposite. You don't get to pick the framework. But you do get to decide how ready your data is for whichever one wins.

🧑‍🏫 PI Presentation Intelligence - Ai-Native Design & Sharing Platform, generate presentations in seconds.

✍️ SEO.ing - Your Go-To Solution for Quick, Effective, and SEO-Optimized Article Writing

🤖 Gumloop - Automate any workflow with AI.

⚙️ Agent Minimax - Research, PPT, Code and more with Minimax . A global AI foundation model company.

🖊️ Scribe by Appblit - Readable transcripts with automated sections and headings.

😍It was great, all I needed

🙂Good, not great

🙄It sucked