AWS Just Started Selling Its Agentic Shopping Stack. It Doesn't Fix Your AI Visibility.

TL;DR: AWS launched the Agentic Shopping Assistant (ASA) on May 27, 2026, packaging the same Claude Haiku 4.5 architecture that powers Amazon's own Alexa for Shopping for outside retailers. Tapestry's Kate Spade is live with a Gift Concierge after a 2.5-month build; deployment runs as little as 60 days. First time a top-three hyperscaler sells agentic shopping infrastructure as a packaged vendor product, and it changes the buyer conversation. But ASA puts an Amazon-powered agent inside your storefront. It does not change how your products surface inside ChatGPT, Gemini, Perplexity, or Google AI Mode. With AI source traffic to retail sites up nearly 400% YoY in Q1 2026, the gap between on-domain agent and off-domain visibility is now the central decision every retailer has to make this quarter.

Until this week the agentic commerce vendor map looked like two camps. The platform giants pitching AI inside the storefront you already run (Salesforce Agentforce, Shopify Sidekick, Microsoft Copilot for Commerce). And the legacy on-site search and discovery specialists rebuilding their stacks for AI. Both targeted the retailer's own website.

On May 27, AWS broke the symmetry. The AWS Agentic Shopping Assistant (ASA) is the Alexa for Shopping stack, the one Amazon uses inside its own walled garden, wrapped as a deployable product for outside retailers. Tapestry's Kate Spade AI Gift Concierge went live April 13 after a 2.5-month build. AWS says deployment runs as little as 60 days, and retailers keep their catalog, customer data, and branding.

A hyperscaler now sells the agent. The harder question is what problem the agent solves, and what problem it does not.

What ASA actually is

ASA is built on Anthropic Claude Haiku 4.5 via Bedrock and AgentCore. It gives a retailer a conversational shopping agent that lives on the storefront, ingests the catalog, talks to customers, and hands transactions to checkout. eMarketer's brief describes the value cleanly: "richer first-party shopper insights, higher conversion rates, more personalized customer engagement." Every phrase there describes work that happens after the shopper has decided to visit the site.

Building a conversational layer like this used to take a year and a team of seven. AWS collapsed it to a quarter and a vendor relationship. For mid-market retailers without an ML team, ASA is useful.

It is also not the gate that determines whether the shopper ever shows up.

The gate that determines whether the shopper shows up is somewhere else

Adobe's data, reported by Adweek on May 28, shows AI source traffic to retail sites climbed close to 400% year-over-year in Q1 2026. The surfaces driving that growth are not retailer storefronts. They are ChatGPT, Google AI Mode, Perplexity, Claude, Gemini, and the assistant embeddings inside Klarna, WhatsApp, and Instagram. When a shopper opens ChatGPT and asks "what's a good gift for a friend who just had a baby," the answer is composed before any retailer in the answer gets a chance to greet that shopper with an in-store agent.

Key stat: AI source traffic to retail sites was up nearly 400% YoY in Q1 2026 (Adobe, via Adweek, May 28, 2026). Automated agent traffic across all sites grew 23.5% in 2025, eight times the rate of human traffic growth (Human Security).

ASA is downstream of that decision. The Kate Spade Gift Concierge is excellent at turning intent into a transaction once the shopper is on katespade.com. It cannot influence whether ChatGPT mentioned Kate Spade at all when the shopper first asked. Two different problems, two different remedies. ASA solves one. It is silent on the other.

Most buyers will conflate them because the marketing language sounds identical: "AI shopping," "agentic commerce," "conversational discovery." The AI visibility problem and the on-domain assistant problem are not the same problem.

The two-surface map every retailer should be drawing this week

A retailer's AI strategy in mid-2026 has to answer two questions independently:

Retailers who buy ASA and assume the second question is now answered will discover in the back half of 2026 that conversion inside their storefront went up while share of voice inside ChatGPT and AI Mode stayed flat. Both numbers matter. Only one will be on the AWS dashboard.

The platform-dependency question

The eMarketer brief raised it gently; the AWS announcement did not address it at all: retailers' "outsized dependencies" on Amazon. The same company offering you its agent is also your largest competitor in product search, the operator whose private-label brands compete on your shelves, and the marketplace that already ingests your catalog. ASA runs on Bedrock and AgentCore. Every conversation a shopper has with your Kate Spade Gift Concierge produces telemetry that lives, by default, inside Amazon's infrastructure.

This is not a reason to refuse the deal. It is a reason to make sure your AI visibility posture, the off-domain visibility layer that determines whether ChatGPT or Gemini mentions your brand, is not also sitting inside the same vendor relationship. Single-vendor lock-in is acceptable for a conversion layer. It is dangerous for a discovery layer where your competitor controls the surface.

What to do in the next 30 days

Three concrete steps for any retailer between $50M and $5B in annual revenue whose CEO just forwarded the AWS news.

1. Separate the two budgets on paper before any vendor conversation. One line for "agent on my storefront," one line for "visibility inside agents I do not own." Decide which is more urgent. For most retailers in apparel, beauty, home, and consumer electronics, the second is more urgent right now because AI-source traffic is already in the high single-digit to low double-digit percentages of referral. The first only matters once shoppers arrive.

2. Audit your catalog readability before you commit to ASA or any peer. ASA reads your catalog. So does ChatGPT's shopping research stack. So does Google's Universal Commerce Protocol implementation. The gaps in attributes, descriptions, and schema that limit your AI visibility also limit your on-domain agent. A bad catalog produces a polite, confidently wrong AI shopping assistant on your site and an invisible product card inside ChatGPT. A structured product data audit is the prerequisite to both bets.

3. Set up share-of-voice tracking across at least four AI surfaces before you launch any on-domain agent. ChatGPT, Google AI Mode, Perplexity, and Claude. You need a baseline before you commit, because in nine months your board will ask whether the investment moved the number. Tools like our AI Readiness Report exist for this; manual sampling works too if you want to start lower-cost.

What this means for the vendor map

Three shifts.

First, hyperscalers are category participants now. AWS sold a product. Google's Universal Cart ships with Nike, Target, Ulta, and Wayfair this summer. Microsoft Copilot for Commerce sits in the same conversation. The "we just sell the cloud" framing is over.

Second, the buyer conversation gets louder, not clearer. A CMO will hear "AWS has an AI shopping product," "the on-site search vendor has an AI shopping product," and "the catalog enrichment startup has an AI shopping product" in one week and assume they are competitive offerings. They are not. Three different surfaces with three different remedies. Anyone selling needs a one-slide answer to "how is yours different from AWS." Honest answer: ASA is the agent on your storefront; AI visibility tooling measures how you appear on the agents you do not own.

Third, catalog quality is now the single point of failure. ASA, UCP, ChatGPT product feeds, Google Merchant Center, and every other surface read from the same source of truth. Digital-shelf audits used to be hygiene. They are now the highest-leverage spend a retailer can make before any vendor commitment.

FAQ

Is AWS ASA a competitor to Salesforce Agentforce or Shopify Sidekick?
Functionally yes. All three are first-party AI agents that live on the retailer's storefront and improve on-domain conversion. ASA's differentiation is the Alexa-for-Shopping lineage and Anthropic Claude Haiku 4.5. Salesforce and Shopify integrate more tightly with their own commerce platforms; ASA with AWS infrastructure. None improves how products surface inside third-party assistants like ChatGPT or Gemini.

Does ASA fix the product card gap inside ChatGPT?
No. ASA operates inside your own storefront. The product card gap, the difference between being mentioned by an AI assistant and being shown as a buyable card, is a function of your catalog structure, your feed presence inside surfaces like the ChatGPT shopping research stack, and your participation in protocols like UCP and ACP. ASA does not touch any of those.

If we use ASA, do we still need an AI visibility strategy?
Yes. ASA improves what happens after a shopper arrives. AI visibility tooling improves whether the shopper arrives at all. With AI-source traffic up nearly 400% year-over-year, the second question is at least as urgent as the first for any retailer whose growth depends on new customer acquisition.

Is Kate Spade's go-live a meaningful proof point?
A useful early signal. 2.5 months from kickoff to live is fast for a custom AI agent. But "live with one customer" is not "proven across the industry." Watch the next three to five retailer go-lives to see whether the 60-day claim holds at scale.

Does using ASA put my catalog data inside Amazon's infrastructure?
Practically, yes. Model calls, conversation logs, and telemetry flow through AWS systems by default. AWS says retailers retain ownership of their data, but ownership and operational location are different questions. Retailers in categories where Amazon competes directly should structure the contract accordingly.

What should we do this week if our board is asking about the AWS news?
Send them the two-surface table above. Make clear that ASA is one half of the answer, off-domain visibility is the other half, and catalog readiness is the prerequisite to both.

The AWS announcement is real news about exactly one of the two surfaces that determine whether a retailer wins in AI shopping. Treat it that way.