Anthropic has released "Claude for Commerce Agents," a blueprint designed to help engineering teams build and deploy retail AI agents in a matter of days. The launch provides the patterns, harnesses, and guardrails necessary to integrate Claude into retail, travel, telecom, and ticketing platforms.
According to the announcement, retailers running shopping agents on Claude have seen carts up to 35% larger, with shoppers 60% more likely to complete a purchase. Enterprise companies like Shopify and Priceline are already using Claude to build systems that allow consumers to search, compare, and buy using plain language. This new blueprint makes the underlying architecture for those systems broadly available, deploying across platforms where businesses already build, including the Claude API, Amazon Bedrock, Microsoft Foundry, and Google Cloud Vertex AI.
The Two Agent Framework
The repository provides working reference implementations for two distinct types of commerce agents: one for the customer, and one for the operator.
The Shopping Agent Designed to live inside an app or website, the shopping agent handles customer-facing interactions. It integrates directly with a business's existing catalog, cart, customer preferences, and order history, while leaving the actual payment processing to the business's existing checkout provider.
Instead of simple keyword searches, a customer can submit multi-item requests, such as asking for a tent, sleeping bag, and stove for a specific type of trip. The agent searches the catalog, assembles the items, remembers user preferences, and displays the products and cart directly within the conversation. It can also handle customer service inquiries in the same thread—such as explaining refund policies or tracking an order—eliminating the need to route customers to a separate support page. Crucially, the blueprint includes guardrails designed to restrict the agent to actual catalog data and pricing, avoiding manipulative upsell tactics or hallucinated discounts.
The Merchant Agent Built for the people running the store, the merchant agent acts as an internal operations assistant. It connects to store data to answer questions about sales performance and track inventory. It can proactively flag issues, such as an item that is on pace to sell out before a scheduled promotion begins.
It can also recommend pricing adjustments based on historical sales data and draft marketing campaigns to help move stagnant inventory. To maintain control, Anthropic designed the merchant agent so that any proactive changes it suggests require human approval before going live.
What This Means for SMB Operations
For small and mid-sized businesses, the barrier to entry for agentic commerce just dropped significantly. Until now, deploying an AI that could reliably interact with your product catalog, handle multi-step customer inquiries, and interface with backend inventory data was an enterprise-level engineering challenge. This blueprint provides a foundational starting point, allowing smaller development teams and automation partners to implement complex functionality without starting from scratch.
While the headline metrics focus on larger cart sizes and customer conversion rates, the operational capabilities of the Merchant Agent are where SMBs stand to gain immediate efficiency. Managing inventory thresholds, analyzing sales trends to clear out last season's stock, and spinning up corresponding marketing campaigns usually requires hours of administrative work from staff who are already stretched thin. By automating the initial analysis and data gathering, businesses can turn complex administrative chores into a quick approval workflow.
Finally, the explicit focus on safety guardrails makes this a practical tool for smaller operators. SMBs cannot afford the financial or reputational damage of an AI hallucinating a 90% discount or offering products that do not exist. By constraining the shopping agent to real catalog data and enforcing a human-in-the-loop approval process for internal merchant actions, the blueprint aligns with how businesses actually need to deploy AI: as a highly capable assistant, not an autonomous, unsupervised manager.