Amazon Web Services just started selling retailers a productised version of the AI shopping technology behind Alexa for Shopping. Starter code, recommendation architecture, operational playbooks — the whole package, deployable in sixty days. Kate Spade is the first named customer. Others are reportedly testing it behind closed doors.

The pitch came with a piece of advice that made me read the blog post twice: retailers should build their own AI shopping tools rather than hand control of the customer relationship to "an intermediary."

Which is a perfectly reasonable argument. It is also, and I cannot stress this enough, coming from the company that built Buy for Me — an AI agent designed specifically to go onto other retailers' websites and complete purchases without those retailers' involvement.

That's like your neighbour smashing your garden fence, then knocking on your door to sell you a home security system. While still holding the sledgehammer.

Here's the thing. Amazon isn't uniquely hypocritical here — they're just the most brazen about it. Every major platform in AI commerce is currently playing the exact same game: trying to be the agent that sits between shoppers and products while simultaneously fighting off every other company trying to do the same thing to them.

OpenAI built ChatGPT Shopping to mediate between consumers and retailers. Google launched its own Buy for Me agent. Perplexity bolted shopping features onto its search product. All three positioned themselves as the intelligence layer between a shopper and a catalogue. And in every case, the retailers on the receiving end had the same delightful choice: become inventory for someone else's AI, or build your own.

Most chose both. Walmart, Target, Etsy, Gap, and eBay have all partnered with OpenAI or Google while simultaneously developing their own proprietary AI shopping tools. Which is the corporate equivalent of dating someone while also maintaining an active profile on three other apps. (Not judging. We've all hedged our bets.)

Shopify took the most diplomatically passive-aggressive approach I've seen in enterprise software. They built Agentic Storefronts to let merchants feed structured product data into AI systems, then added explicit language to their robots.txt warning against automated agents completing checkout without human review. Translation: we'll let your AI browse, but it doesn't get to swipe the credit card. The digital equivalent of "you can look at the menu, but I'm ordering for myself, thanks."

Amazon, characteristically, chose the nuclear option. Blocked external agents. Built its own outbound agent. And is now selling the defensive tools to everyone else.

AWS framed this as business as usual. Amazon builds technology internally, solves an operational problem, then commercialises it as a service. AWS itself is the textbook example. Cashier-less checkout (Just Walk Out), warehousing robotics, supply chain logistics — same pattern every time. Build it, use it, sell it.

But there's a meaningful difference between selling someone cloud computing capacity and selling them a commerce intelligence agent. AWS is plumbing. It's deliberately boring infrastructure. A retailer running on AWS isn't importing Amazon's retail brain into its business. The pipes don't have opinions about which products to recommend.

An AI shopping agent built on Alexa for Shopping's architecture? That's a different animal entirely. This thing packages the recommendation logic, conversational design patterns, and product understanding frameworks that Amazon developed to serve its own commercial interests. It's not plumbing. It's the plumber, and the plumber has strong views about which taps you should install.

The fact that it's being offered through AWS rather than Amazon's retail division is clearly intended to create separation. AWS has spent years building trust with enterprise customers who compete directly with Amazon's retail arm, and routing the product through that channel signals that data governance follows AWS norms rather than retail ones.

Whether that distinction holds when things get competitive — particularly as Amazon keeps expanding its own agent capabilities — is a question that the announcement conveniently did not address.

The most remarkable thing about this announcement isn't the product. It's the argument.

Amazon's blog post explicitly warned retailers against surrendering the shopping experience to intermediaries. Build your own tools. Control your own data. Leverage your vertical expertise.

This is the same company whose Buy for Me agent browses third-party retailer websites, selects products, enters payment information, and completes purchases — all without the retailer's direct involvement. The shopper never leaves Amazon's interface. The retailer fulfils the order but has zero relationship with the customer. When Buy for Me first launched, small retailers discovered Amazon's agent was purchasing from their sites without consent, creating customer service nightmares around returns and support.

And — and this part is genuinely exquisite — Amazon blocked Google's shopping agent from accessing its own platform while simultaneously scraping everyone else's.

So to recap the current state of play: Amazon is warning retailers about the dangers of intermediaries. Amazon is selling them tools to defend against intermediaries. And Amazon is the intermediary they most need defending against.

It's a business model so circular it could be classified as a roundabout.

Here's where it gets uncomfortable. Strip away the source, and Amazon's argument is actually... correct?

The AI shopping landscape is fragmenting fast. Retailers relying entirely on external platforms for AI-mediated discovery are ceding strategic ground with every passing quarter. OpenAI's retreat from Instant Checkout earlier this year proved that owning the checkout experience still matters. Walmart's data showing that purchases completed inside ChatGPT converted at roughly one-third the rate of those completed on its own site reinforced the point. Discovery may be migrating to AI interfaces, but conversion still lives where the merchant controls the experience.

For mid-market retailers without the engineering resources to build a proprietary commerce agent from scratch — and that's most of them — the AWS offering fills a real gap. Salesforce has been pitching its own agent-building services. Shopify's infrastructure serves its merchant base. But for retailers outside those ecosystems, particularly those running on legacy platforms held together with digital duct tape and good intentions, the options have been "partner with an AI platform and become inventory" or "do nothing and hope for the best."

A sixty-day deployment timeline, backed by AWS infrastructure and built on a recommendation engine that has processed hundreds of millions of shopping interactions, is a genuinely credible product.

The question isn't whether it works. The question is whether a retailer adopting Amazon's commerce intelligence as the foundation of its own AI shopping experience is building independence or a more sophisticated form of dependence. (My money is on "sophisticated dependence," but I'd love to be wrong.)

What Amazon's announcement crystallises is a tension running through every development in agentic commerce this year. The industry is hurtling toward a structural collision where the roles of agent and storefront are being contested by every single participant simultaneously.

Amazon wants to be the agent (Buy for Me), the infrastructure provider (AWS), and protect its own storefront from other agents. Google wants to be the agent (Buy for Me, AI Mode shopping) while maintaining advertising relationships with the retailers those agents disintermediate. OpenAI wants to mediate discovery while retreating from transaction ownership. Shopify wants to be the middleware layer feeding all agents while preventing any single agent from owning checkout. Walmart wants to participate in AI discovery channels while keeping its own branded agent as the customer interface.

Everyone is trying to be the intelligence layer. Nobody wants to be the passive catalogue that someone else's intelligence layer queries. It's like a room full of people all trying to be the DJ and nobody willing to dance.

This is not a stable situation. Someone has to be the storefront. The economics of agentic commerce require products to exist somewhere, with real inventory, real fulfilment, and real customer service. The question is whether the companies that own that infrastructure can maintain a direct relationship with the shopper, or whether the agent — whichever agent wins — becomes the de facto customer.

Amazon selling AI shopping tools to retailers is either a genuine effort to distribute technology that keeps merchants in control of their own customer relationships, or a play to ensure that whichever agent a retailer builds, it runs on Amazon's infrastructure, uses Amazon's architecture, and feeds data through Amazon's cloud.

Both of those things can be true at the same time. And that's the nature of this entire phase of commerce: in a market where everyone is trying to be the agent, the company selling the tools to build agents might end up with more power than the agents themselves.

Amazon looked at the agentic commerce landscape, saw that everyone was scrambling to build defensive walls, and decided the smartest move was selling bricks. The fact that they're also the one with the battering ram? That's not a bug. That's the business model.

P.S. If you're a seller trying to figure out where your storefront fits in all of this, the priority hasn't changed: own your customer data, control your checkout experience, and treat every platform partnership as a rental agreement, not a mortgage. The landlord's incentives and yours will diverge eventually. They always do.

🪢 Zapier - Connect thousands of apps and add AI steps to automate workflows.

🌐 n8n - Open-source, self-hostable workflow automation with custom code nodes.

🤖 Decagon - AI support agents that resolve refunds, account and order tasks end to end.

🪄 Magic Patterns - Generate and prototype UI designs and components from a prompt.

🏄‍♀️ Surfer SEO - Analyze top-ranking pages and optimize content to rank higher.

😍It was great, all I needed

🙂Good, not great

🙄It sucked