Anthropic publishes blueprints for shopping and merchant agents

What changed

Anthropic's release provides reference implementations for two different commerce tasks. The shopping agent helps a customer explore a catalogue, assemble a cart and move to the merchant's checkout. The merchant agent supports store operations such as sales analysis, inventory review and proposed promotions. The examples span retail, travel, telecoms and ticketing. They are starting points for integration, rather than a single finished storefront that every advertiser can switch on.

The accompanying engineering guide is written by Matthew Koen and Ali Shazal. It explains how the model's tools should connect to existing business systems, with proposed commercial changes checked and approved through the surrounding application. That separation is useful for a retail-media team: an agent might identify slow-moving stock and draft a promotion, while the business still needs to decide whether the proposed discount, stock position and campaign budget fit its plan.

A sensible pilot would follow one customer or merchant journey from beginning to end. For a shopper, that might mean finding an in-stock product that satisfies a stated constraint and reaches the correct cart. For the merchant, it might mean turning a stock issue into a reviewable campaign proposal. These are suggested evaluations, not evidence that the supplied blueprint has already been validated against the reader's catalogue, transaction rules or customer experience.

Why it matters for advertisers

The opportunity is to connect discovery and operational analysis more closely with commerce data. Paid-media teams can contribute the real questions, product claims and promotion constraints an engineering pilot must handle. Judge success by correct completed tasks and useful commercial outcomes, including how much review or repair is needed, rather than adopting the announcement's customer-success figures as a forecast.