Anthropic Launches Commerce-Agent Blueprints

Anthropic has released commerce-agent blueprints that businesses can use to build customer-facing shopping tools and internal merchant assistants with Claude. The resources include working reference implementations for retail, travel, telecom, and ticketing businesses.

The shopping agent can search a company’s catalog, compare products, assemble a cart, and pass the customer to checkout. A separate merchant agent can analyze sales, monitor inventory, and recommend pricing or promotional actions. The release does not include an affiliate attribution or commission system, although retailer-controlled product discovery raises questions for programs that rely on external publishers and creators to introduce customers.

Key Takeaways: Anthropic’s Commerce-Agent Blueprints

• Anthropic has released working reference implementations for shopping and merchant agents.
• Businesses can customize the blueprints using their own catalogs, policies, and operational data.
• The shopping agent operates inside the company’s website or app and connects with its existing checkout.
• The merchant agent can analyze sales, monitor stock, and recommend pricing or marketing actions.
• Anthropic says people retain approval over operational changes proposed by the merchant agent.
• No dedicated affiliate tracking, attribution, or commission functionality has been announced.

What Anthropic Has Released

According to the official Anthropic announcement, the blueprint contains the technical framework, integration patterns and safeguards required to build commerce agents using Claude. It also includes a Claude Code plugin and self-guided demonstrations for the supported industries.

Businesses can deploy the blueprints through the Claude API, Amazon Bedrock, Microsoft Foundry or Google Cloud Vertex AI. Anthropic is also working with companies including Accenture, Mastercard and Visa to help businesses implement the technology.

The release contains two principal reference implementations. The shopping agent is designed for customers researching and purchasing products or services. The merchant agent supports the teams managing the commercial operation.

The underlying code is available for companies to adapt to their own catalogs, customer policies and internal systems. Anthropic is therefore providing the foundation for businesses to build their own agents rather than releasing a single consumer shopping destination.

The Shopping Agent Connects Product Research With Checkout

The shopping agent operates inside a retailer’s website or app. It includes connections for product catalogs, carts, checkout systems, customer preferences and order histories.

A customer can describe what they need in conversational language rather than entering individual product searches. The agent can then:

• Search the merchant’s catalog.
• Compare suitable products.
• Respond to requests involving several items.
• Remember customer preferences.
• Display products and comparisons within the conversation.
• Build a cart and pass it to checkout.
• Answer questions about delivery, returns, exchanges and refunds.

Payment remains with the merchant’s existing checkout or a separate agentic-payment provider. Anthropic does not become the merchant of record or process the purchase through a centralized Claude marketplace.

This separates the blueprint from agentic-commerce systems that allow discovery and payment to happen entirely inside an external AI platform. Affiverse has previously examined how Google and Shopify’s Universal Commerce Protocol connects AI product discovery with checkout. Anthropic’s approach instead gives individual businesses a way to place similar conversational tools inside their own digital properties.

The Merchant Agent Supports Internal Commercial Decisions

The second blueprint is intended for the people running the business rather than its customers.

The merchant agent can answer questions about sales performance, identify products that are selling slowly and flag stock problems before they affect planned promotions. It can also recommend discounts based on historical sales information and draft marketing campaigns for products the merchant wants to move.

Anthropic says operational changes require human approval before they are implemented. The agent can identify a potential action and prepare a recommendation, but a person retains the final decision.

That distinction is important because pricing, inventory and promotional decisions can affect customer expectations, campaign terms and existing partnerships. The blueprint is designed to support those decisions rather than give the agent unrestricted authority to change the commercial operation.

Anthropic Reports Higher Cart and Conversion Results

Anthropic says retailers using shopping agents built with Claude have recorded carts up to 35% larger and shoppers who were 60% more likely to complete a purchase.

The company told Reuters that the figures came from one partner. They should therefore be treated as an example supplied by Anthropic rather than a benchmark for commerce agents generally.

The announcement does not identify the partner, provide the underlying sample size or explain how the results were measured. It also does not establish whether equivalent improvements would occur across different merchants, product categories, or customer groups.

Performance will depend on factors including catalog quality, recommendation accuracy, pricing, stock availability, checkout design, and the type of purchase being considered.

Merchant-Owned Agents Raise Product-Discovery Questions

Although the agent operates on the merchant’s property, it could change the stage at which the final product decision occurs.

A customer may arrive after reading a publisher’s comparison or watching a creator demonstrate a particular item. Once on the retailer’s website, the customer could ask the agent to compare alternatives, select another model or assemble a larger bundle.

The affiliate may still have introduced the customer, while the merchant’s agent determines which products ultimately enter the basket. Whether the original referral receives credit will depend on the program’s tracking setup and commission rules.

The presence of an agent does not automatically remove an affiliate cookie or click identifier. If the customer enters through a tracked link and completes checkout within the same recognized journey, conventional attribution may continue to operate.

Questions may arise when the recommended product changes, additional items carry different commission rates or the customer returns through another device or session. These issues resemble the wider challenges already seen with cross-network affiliate commission deduplication, where separate systems can hold incomplete views of the same transaction.

What Retailers and Affiliate Teams Still Need to Establish

The blueprints demonstrate what commerce agents can do, but retailers must still prove that customers trust their recommendations and that the technology creates value beyond simplifying an existing purchase journey.

Agents may work best for products that can be assessed through specifications, availability and price. Subjective categories such as clothing, gifts and beauty may be more difficult. A merchant-owned agent is also limited to its retailer’s catalog, making transparency around product selection important.

Key questions include:

1. Can the agent explain why it selected a product and identify promotional influences?
2. Does it accurately reflect prices, stock, and product variants?
3. Can customers correct remembered preferences and approve the final cart?
4. Does the original affiliate identifier survive the agent-led session?
5. How are substitutions, additional items, and agent-assisted conversions treated for commission?

Anthropic’s reported results do not show whether larger baskets came from the agent or customers who were already likely to purchase. These questions connect with work on open attribution for AI-assisted customer journeys, which aims to preserve information about earlier content influence.

An agent does not automatically remove affiliate tracking. Its effect will depend on the retailer’s implementation, program rules, and ability to show what the agent contributed while continuing to recognize the partner that introduced the customer.