Anthropic ships blueprints for Claude shopping agents that buy on your behalf

Anthropic this week published a set of templates on GitHub for building Claude-based e-commerce agents: a shopping agent that can search, compare and buy on a customer's behalf, and a merchant agent that businesses can run on their own side. The blueprint can be built with the Messages API, the Agent SDK or Claude Managed Agents, and Anthropic says it "contains the harnesses, patterns, and guardrails an engineering team needs to get a commerce agent running in days," with reference implementations for retail, travel, telecom and ticketing platforms. The shopping agent connects to product catalogs, shopping carts, checkout systems, preference databases and purchase history, and Anthropic gives the example of a customer saying "I need a tent, sleeping bag, and stove for a weekend trip with two kids," leaving the agent to handle the rest. The company says the code includes guardrails "designed to constrain prices and products to actual catalog data, and avoids manipulative upsell patterns."

The push runs ahead of consumer appetite. A recent Gartner survey found that while people use AI to research and compare products, just 11 percent are willing to let AI make the actual purchase decision. An Accenture survey paints a somewhat friendlier picture: 74 percent of consumers would let an AI agent handle routine tasks, 32 percent are ready to hand over purchase decisions, and 9 percent say they are open to fully autonomous shopping even before the technology is fully operational.

The article also raises a pricing worry that predates Anthropic's release. The Brookings Institution warned earlier this year that agentic AI, which can monitor online behavior to customize costs, is likely to worsen dynamic pricing, since such agents can infer a shopper's history and tailor the price shown to them. That concern surfaced last month at a Senate Judiciary subcommittee hearing on AI surveillance pricing, where Lindsay Owens, president and CEO of the Washington-based advocacy group Groundwork Collaborative, testified that half of Walmart app users already use its Sparky AI assistant, and that those shoppers spend about 35 percent more than non-users. Owens said such agents have access to data ranging from what a shopper tells them to what they have purchased before and even what they have hovered over with a mouse, all of which can be used to infer how much a customer will pay.

A second concern is fraud liability. Monica Eaton, founder and CEO of Chargebacks911, told The Register that agentic commerce looks likely to create fraud problems for merchants, since there is not yet a framework for handling AI purchases that a consumer later claims were unauthorized. "Merchants cannot become the insurer of every misunderstanding between a consumer and their AI," she said. The piece concludes that Anthropic's templates may help on the engineering side, but the harder problem is trust: consumers need to trust that an agent acts in their interest, and merchants need to trust that it will be accountable.

Key facts

Anthropic published GitHub blueprints for Claude-based shopping agents (for consumers) and merchant agents (for businesses), buildable via the Messages API, Agent SDK or Claude Managed Agents, with reference implementations for retail, travel, telecom and ticketing.
A Gartner survey found only 11 percent of consumers are willing to let AI make purchase decisions, even though they use AI for product research.
An Accenture survey found 74 percent would let an AI agent handle routine tasks, 32 percent are ready to hand over purchase decisions, and 9 percent are open to fully autonomous shopping already.
Groundwork Collaborative's Lindsay Owens told a Senate Judiciary subcommittee that half of Walmart app users use its Sparky AI assistant, and those users spend about 35 percent more than non-users, raising dynamic-pricing concerns flagged earlier by the Brookings Institution.
Chargebacks911 CEO Monica Eaton warns agentic commerce lacks a fraud framework for purchases consumers later claim they did not authorize.

Why it matters

Anthropic is trying to move agentic commerce from demo to default by handing engineering teams a ready-made harness rather than waiting for businesses to build shopping agents from scratch. That is a bet that the infrastructure, not consumer demand, has been the bottleneck.

Who it affects

Retailers and e-commerce platforms deciding whether to deploy AI shopping agents, the consumers who would interact with them, and merchants who would need to build the corresponding merchant-side agent and absorb any fraud risk from AI-initiated purchases.

How to use it

The templates are available now on GitHub and can be built with Anthropic's Messages API, Agent SDK or Claude Managed Agents; reference implementations cover retail, travel, telecom and ticketing use cases, with connectors for catalogs, carts, checkout, preferences and purchase history. No pricing or licensing terms for the templates are given in the source.

How solid is it

The release itself is confirmed by Anthropic's own description of the blueprint, and the surveys cited (Gartner, Accenture) and testimony (Owens, Eaton) are attributed to named sources. The article does not name any retailer or platform already using the templates, only describing them as newly available.

Risks and caveats

Consumer willingness remains low by Gartner's measure, so adoption is not guaranteed. Dynamic pricing driven by agents that can infer a shopper's willingness to pay is flagged by both Brookings and Owens as a live concern, and Chargebacks911's Eaton points to an unresolved gap in fraud liability when a consumer disputes an AI-made purchase.

“Merchants cannot become the insurer of every misunderstanding between a consumer and their AI”

— Monica Eaton, founder and CEO of Chargebacks911