Anthropic wants to turn the holiday shopping rush into the moment AI agents finally earn their keep in retail. On Wednesday the company released a set of open blueprints for building shopping and merchant agents on Claude, and it is pitching them straight at retailers weeks before the busiest quarter of the year.
The timing is deliberate. Shoppers already lean on AI tools to compare prices, check whether a product is in stock, and collect recommendations before they click buy. Adobe Analytics reported last month that site visits arriving through AI-driven assistants convert at a roughly 60 percent higher rate than traffic from any other source. Anthropic's pitch is that the pieces it is handing out let a store get one of those agents running in days rather than months.
The blueprints are reference implementations, not marketing fluff. They contain the patterns, guardrails, and reusable code an engineering team needs to stand up two kinds of agents: one for the shopper and one for the person running the store. Both come ready for retail, travel, telecom, and ticketing businesses, and both ship with a Claude Code plugin so a developer can drop them into an existing workflow.
Deployment is deliberately open-ended. The code runs wherever teams already build on Claude, whether that is the Claude API, Amazon Bedrock, Microsoft Foundry, or Google Cloud Vertex AI. Anthropic is also leaning on its solutions and ecosystem partners, among them Accenture, Mastercard, and Visa, to help merchants put the tools to work.
What the shopping agent actually does
The shopping agent lives inside a brand's app or website. It plugs into the catalog, cart, checkout, customer preferences, and order history, but leaves payment handling to the merchant, whether that is the store's existing checkout or an agentic payments provider. Anthropic leaves payments out of the reference code on purpose, because a shop should get to choose how money moves.
A customer could type a natural request such as "I need a tent, sleeping bag, and stove for a weekend trip with two kids," and the agent takes it from there. It can search the catalog, plan a multi-item trip, research options, personalize what it shows, and handle follow-up questions in the same conversation. The blueprint ships with guardrails built in: it constrains prices and products to the actual catalog data and avoids manipulative upsell patterns.
Anthropic said retailers already running Claude-based shopping agents are seeing real results. Angela Jiang, the company's head of product for the Claude platform, pointed to partners reporting cart sizes up to 30 to 35 percent larger and customers about 60 percent more likely to finish a purchase. Those are the kinds of numbers a retailer can take to a planning meeting.
Anthropic's own announcement led with the same figures. Retailers running shopping agents on Claude have seen carts up to 35 percent larger and shoppers 60 percent more likely to complete a purchase, the company said.
The merchant agent keeps a human in the loop
The second blueprint targets the people who actually run a store. A merchant can ask something like "what should we discount to clear last season's inventory?" and get an answer built from its own sales data rather than a generic suggestion.
The merchant agent comes with skills for sales analytics, catalog and inventory management, marketing and promotions, plus in-portal dashboards and charts. The important part is the approval flow: when the agent proactively suggests a change, a person has to approve it before anything goes live. The machine watches the store, but the human keeps the final say.
That distinction matters to Anthropic as it courts cautious buyers. The company has spent the past year positioning Claude around safety and control, and the commerce push leans on that reputation. A checkout agent that cannot ring up a sale on its own, and a store-operations agent that cannot move a price without sign-off, are both easier for a retailer to defend to its security and compliance teams.
The move also shows how Anthropic intends to compete for real-world business rather than just developer mindshare. Shopping and commerce are concrete, measurable use cases: an agent either lifts the average cart or it does not, and a retailer can judge the ROI in a single quarter.
Who is already building with it
Anthropic is not starting from zero. The announcement lists established customers already running Claude-powered agents, including Shopify, Priceline, Intuit, Wix, Klaviyo, Square, and Zomato. Several shared early results.
Wix said its engineers had a working commerce agent taking prompts within fifteen minutes, and that a pilot showed what pairing Claude's abilities with the Wix platform could do for small and mid-size merchants. Zomato reported that its teams got the blueprint running with no blockers, and that the practices baked into it, from tool iteration limits to prompt caching, matched what Zomato had learned building its own agent.
Accenture contributed a survey angle. Its latest research found that 85 percent of consumers are now open to working with an AI agent, and that nearly three in four would trust a personal AI agent more than their best friend to make a purchase on their behalf. Those are striking numbers, even allowing for a consulting firm's enthusiasm about the market it serves.
The blueprint is available now, with live demos for each vertical and an engineering deep-dive that explains how it was built. Anthropic is scheduling the launch just in time for holiday season planning, when retailers are deciding which projects get budget for the fourth quarter.
For the folks betting that AI commerce stays a demo, this is a reminder that the infrastructure is already in place. Anthropic's blueprints can be rebuilt from source, deployed on any major cloud, and wired into a store's existing checkout. Whether that turns into a measurable bump in fourth-quarter sales is the test that matters.
Anthropic frames this as the next chapter of conversational commerce, the kind that started with chatbots on store sites and is now moving toward agents that can reason across a catalog and act on a customer's behalf. The company has said before that it wants Claude to be the model businesses trust for real transactions, and the commerce blueprints are the most direct bid yet to make that happen.
The road is not entirely smooth. Consumer trust in AI agents remains fragile, surveys show, and a single widely publicized shopping-agent mistake could set the category back. That is likely why Anthropic ships guardrails rather than autonomy, and why the merchant agents stop at suggestions until a human clicks approve.
None of this needs a new model release. The blueprints run on today's Claude models and today's APIs, which makes the product available to any retailer tomorrow rather than whenever the next frontier model ships. For a company that has spent months in a crowded agent race, that is a deliberate choice to compete on utility and trust rather than raw capability.
The holiday season will be the trial. If carts grow and completion rates climb for early adopters, expect rivals to copy the playbook quickly. If not, retailers will have the evidence they need to park the idea. Either way, Anthropic has put a stake in the ground for agentic commerce months before the biggest shopping quarter of the year.
The competitive pressure explains some of the urgency. OpenAI, Google, and Microsoft have all been pushing agentic tools at enterprise buyers, and the e-commerce angle is one Anthropic can defend with hard numbers rather than promises. By opening with a working reference implementation and named customers, it is trying to set the terms of the debate: build on open patterns, keep a human in the loop, and measure the result in cart size and completion rate.
For more on how AI agents are reshaping business workflows, see the AI section of this site. Anthropic detailed the launch on its own commerce agents announcement, and Reuters covered the release in depth via Channel NewsAsia and the Economic Times.