Anthropic has released a new “commerce agent blueprint” that enables retailers, marketplaces, and e-commerce platforms to develop their own AI-powered commerce agents on top of Claude models. The ready-made architecture offers two core applications: a shopping agent for consumers and a merchant agent that supports business operations. The company states that engineering teams can deploy the systems within a few days thanks to the required tools, workflow patterns, and safety measures.
Anthropic Brings the Shopping Agent to E-Commerce Websites
The customer-focused AI agent operates within the retailer’s app or website. Users can describe their needs in natural language; the system can find products based on the store’s product catalog, customer preferences, and past orders, compare options, and create a shopping cart. The agent can also answer customer service questions related to order tracking, returns, exchanges, and refund policies within the same conversation.
Consumer data regarding the use of artificial intelligence in retail stands out. In a survey conducted by PSE Consulting with 4,250 consumers in the United Kingdom, the United States, France, and Germany, 74% of respondents preferred an independent AI assistant for shopping, while only 10% preferred an artificial intelligence solution tied to a single retailer or delivery platform. In a separate study, 63% of consumers were reported to be uncomfortable with AI using their browsing and purchase history to generate recommendations.
Merchant Agent Can Monitor Inventory, Pricing, and Campaigns
The second system developed by Anthropic is positioned as a “merchant agent” for employees managing e-commerce operations. The system can analyze sales performance, monitor inventory levels, prepare pricing and promotion recommendations, and create marketing campaigns using company data. According to data released by the company, retailers using Claude-based shopping agents saw basket sizes increase by up to 35%, while the likelihood of completing a purchase rose by up to 60%.
Human Approval and Safety Controls Remain in Place
Anthropic requires human approval before changes recommended by the merchant agent are implemented. The system also includes controls designed to ensure that pricing and product information remain consistent with actual catalog data and to prevent manipulative upselling practices.
Developers can use the solution through the Claude API, Amazon Bedrock, Microsoft Foundry, or Google Cloud Vertex AI. Anthropic is also working with Accenture, Mastercard, and Visa, while Shopify is developing a sample store based on the new architecture; Priceline is using Claude in the latest version of its AI travel assistant, Penny. The blueprint’s code, live demos, and technical documentation are available on GitHub.