This week in AI commerce: AWS rents out its shopping agent, Salesforce ships its May release

Five months into the year, the agentic commerce infrastructure looks less like a race and more like a buildout. The headline this week: AWS turned Amazon’s internal shopping agent into a product anyone can license, with Kate Spade as the first named customer. Salesforce shipped its May 2026 B2C Commerce release alongside the ongoing Agentforce Commerce rollout. And Klarna’s MCP-powered ChatGPT shopping app, launched the prior week, is still doing the rounds.

Here’s what landed between May 25 and May 31, 2026.

---

AWS rents out its shopping agent

AWS Agentic Shopping Assistant launched May 27

AWS announced the Agentic Shopping Assistant on May 27, packaging the technology behind Amazon.com’s Alexa for Shopping into a solution non-Amazon retailers can deploy on their own properties.

The components, per SiliconANGLE’s reporting:

• Amazon Bedrock — foundation model access
• Amazon Bedrock AgentCore — agent development toolkit
• Amazon OpenSearch — managed search over the retailer’s product catalog
• Professional services — the AWS Generative AI Innovation Center plus systems integrators handle deployment

AWS frames the timeline as roughly 60 days to stand up a custom deployment, versus the months-to-years a retailer would otherwise spend building the same stack in-house. Retailers keep their data, their catalog, and their brand rules; AWS provides the agent runtime and the integration scaffolding.

The named launch customer is Kate Spade (parent company Tapestry), which deployed the Kate Spade AI Gift Concierge in April. From AWS: it was the first production retail assistant built with Bedrock AgentCore. Quoted from Yang Lu, Chief Information and Digital Officer at Tapestry: “We are excited about the possibilities agentic commerce can bring to our customers. AWS brought the recipe, but together we built the customization our consumers needed.”

The strategic read: Amazon has been the holdout in the platform-agent productization race. Two weeks ago, the retirement of Rufus in favor of Alexa for Shopping consolidated Amazon’s consumer-facing AI. This week’s announcement does the cloud-side equivalent — turning the underlying capability into something other retailers can pay AWS to operate. It’s the same pattern AWS has run for two decades: build it for Amazon first, productize it once it works.

For brand operators, the practical takeaway is short. The on-site agent — the conversational helper that lives in the merchant’s own storefront — is now a thing you can rent. Whether you should build, buy, or wait is a separate question. But the cost of standing up some version of one just dropped meaningfully.

---

Salesforce: May 2026 B2C Commerce release

Salesforce published the B2C Commerce May 2026 release covering merchandising, product management, and payment efficiency updates. The release sits inside the broader Agentforce Commerce rollout that Salesforce has been pacing across FY26.

The Agentforce Commerce roadmap, per Salesforce’s published announcement:

• Guided Shopping — generally available
• Agentforce Actions for Merchandising — Winter 2026 (beta)
• Order Routing for Order Management and Data 360 Zero Copy Access for Commerce — Winter 2026 (GA)
• Agentforce Actions for Point-of-Sale — Spring 2026 (pilot)

The framing Salesforce uses is consistent across the rollout: a unified platform for digital commerce, point-of-sale, and order management, with the same agent layer running across every shopping surface. Pandora is the most-cited brand customer in Salesforce’s own Agentforce Commerce materials.

For brands running on Salesforce Commerce Cloud (B2C Commerce), the practical implication is that the Agentforce surfaces are arriving on a predictable cadence, and the catalog-readiness work that determines whether those agents can recommend a given product with confidence is the same catalog-readiness work that determines whether ChatGPT or Google AI Mode can. Whichever surface you’re optimizing for first, the underlying data quality moves the needle on all of them.

If you’re new to the SFCC catalog model and want a structured-data starting point, Lumio’s product schema guide for Salesforce Commerce Cloud walks through SFRA, the composable storefront, and the cartridge pattern for adding JSON-LD to a Commerce Cloud deployment.

---

Also on the radar

Klarna’s ChatGPT Shopping Search app keeps pulling. Launched May 20, the app is powered by Klarna’s Product Search MCP server and surfaces real-time prices, availability, and offers from Klarna’s merchant network — over 1 million merchants across 26 markets per Klarna’s own materials — inside a ChatGPT conversation. Checkout is not in-chat; Klarna redirects to the merchant’s site. The MCP layer is the load-bearing piece: Klarna built one server, ChatGPT plugged in, and the network became addressable to the agent surface.

Shopify Agentic Storefronts remain the default for eligible merchants. Since the Winter ‘26 Edition launch on December 10, 2025, eligible Shopify catalogs syndicate to ChatGPT, Perplexity, Microsoft Copilot, and other agent surfaces — no per-platform integration required on the merchant side. The platform list has grown through Q2.

---

What it all means

Three patterns this week:

1. The on-site agent is becoming a rentable layer. AWS’s announcement matters less for what it does (the underlying Alexa-for-Shopping capability already runs on Amazon.com) and more for who it sells to (everyone but Amazon). Pair it with Salesforce Agentforce on the Commerce Cloud side and Shopify Agentic Storefronts on the Shopify side, and the picture is: three of the largest commerce vendors are each productizing a different slice of “the agent your shopper talks to.” Brands no longer need to decide whether to build one. They need to decide whose to rent.

2. The MCP pattern is quietly winning the catalog-syndication layer. Klarna’s ChatGPT app works because Klarna built an MCP server and OpenAI plugged into it. Anthropic’s MCP launched in late 2024; by mid-2026 it’s the connective tissue between catalogs and agent surfaces — not announced as such, just adopted. If a brand’s catalog isn’t reachable through an MCP server (whether the brand’s own, a vendor’s, or the platform’s), the brand is increasingly invisible to the agent layer.

3. The infrastructure layer keeps consolidating; the brand-data layer still hasn’t. AWS, Salesforce, Shopify, Google, Stripe, Visa — the platforms are converging. What hasn’t converged is the catalog quality on the brand side: the structured data, the identifier coverage, the descriptive depth, the real-time accuracy of prices and inventory that determine whether any of these agents pick a given brand’s product over a competitor’s. Renting an agent doesn’t fix any of that. The brands that close the data gap before the agents start picking favorites are the ones that capture the agentic share. The rest watch their competitors do it.

If you want to know where your product data stands across the surfaces that matter — Google AI Mode, ChatGPT, Perplexity, Amazon’s Alexa for Shopping, the SFCC and Adobe Commerce surfaces, and the rest — get in touch. The AI Readiness Score tells you exactly what to fix first.