AWS will sell you Amazon's shopping agent in 60 days

Amazon spent years building the shopping assistant behind Alexa for Shopping. On September 26 it started selling that recipe to everyone else. The AWS Agentic Shopping Assistant is architecture guidance, starter code, and hands-on help from AWS's Generative AI Innovation Center, and AWS says a retailer can be live in roughly 60 days. The build just stopped being the hard part. Your product data still is.

What actually happened

Per AWS's announcement and the AWS industries blog:

• The stack is named. Amazon Bedrock, Bedrock AgentCore, and OpenSearch. Kate Spade's AI Gift Concierge — the first production build, launched April 13 — runs on Anthropic's Haiku 4.5, with AgentCore handling observability, authentication, and evaluations. Tapestry's team ran about 2.5 months of testing before it faced a customer.
• Sixty days, with a caveat. That figure assumes guidance from AWS's Generative AI Innovation Center. AWS frames it as "weeks rather than the years it would take starting from scratch."
• A Shopify connector via UCP. The newest release pulls catalog data through the Universal Commerce Protocol with no custom API mapping. That is the integration tax dropping to near zero for Shopify merchants.
• Six more new features: a self-service admin portal for brand voice and catalog changes without a developer, long-term memory across sessions through AgentCore, multimodal and voice-to-voice search, session history, cart awareness inside the conversation, and conversational intelligence dashboards showing demand signals and trending categories.
• The numbers are Amazon's own. AWS says conversational sessions convert at 3.5x keyword search, that over 300 million customers used Alexa for Shopping last year, and that it drove nearly $12 billion in incremental sales.

Yang Lu, Tapestry's chief information and digital officer, put the division of labor plainly: "AWS brought the recipe, but together we built the customization our consumers needed."

Why agentic commerce tooling matters for your business

Read the 3.5x carefully. It was measured on Amazon's traffic, against Amazon's keyword search, on Amazon's catalog — the best-structured product data on the internet. It is a ceiling, not a forecast. Nobody is handing you that multiple because you shipped a chat widget.

Here is the part the press release does not say out loud: a shopping agent is a reasoning layer over your catalog, and it inherits every defect in it. Missing dimensions, inconsistent color names, variants modeled as separate products, specs living in a PDF nobody parsed, return policy that exists only in a support macro. The agent does not fix those. It surfaces them, confidently, to a customer who is about to buy. We have cleaned up enough product feeds to know the failure mode: the model answers the question your data answers, not the question the shopper asked.

The Shopify-over-UCP connector is the genuinely important line item. When the plumbing is free, the differentiator moves entirely to data quality and business rules — which is exactly where a three-person shop can beat a much larger competitor, because attribute hygiene is work, not budget.

Two things to settle before you sign. First, AWS says you keep control of customer data, catalog, and business rules; get that in the contract rather than the blog post, and confirm where conversation transcripts live and how long they are kept. Second, keep your catalog canonical somewhere you own. If the enriched, agent-ready version of your product data only exists inside the assistant's index, you have rebuilt vendor lock-in with extra steps.

Sources: About Amazon: AWS Agentic Shopping Assistant, AWS Industries Blog: seven new features, GeekWire.