Agent Horizon

Anthropic has released Claude Commerce Agents, an open, Apache-licensed blueprint of code for building AI shopping and merchant assistants on top of Claude. Announced on the official Claude blog, the release ships two reference agents — a shopping agent that searches catalogs, compares products, and builds a cart, and a merchant agent that handles inventory, pricing, and campaign work — with working versions for retail, travel, telecom, and ticketing.

What’s new here

This isn’t a new model; it’s packaged engineering work. The repository runs on the Claude API, Amazon Bedrock, Microsoft Foundry, or Google Cloud Vertex AI, and ships as a Claude Code plugin, so teams can fork a working implementation instead of building agent scaffolding from scratch. Anthropic says Accenture, Mastercard, and Visa are working with it to bring the framework to their client and merchant networks, while Shopify has published its own public reference implementation connecting the agents to live storefronts, and Priceline has already rebuilt its Penny assistant on Anthropic’s models.

Why it matters

Agentic commerce has quickly become a three-way race: OpenAI has been pushing checkout features inside ChatGPT, and Google has its own shopping agent ambitions tied to Search, but Anthropic’s approach here is notably narrower and more infrastructural. The blueprint deliberately excludes any payment protocol, product catalog, or advertising layer — those pieces stay with retailers and their existing partners, so no transaction actually routes through Anthropic. That’s a sensible line to draw given how unsettled agentic payment standards still are; Visa, Mastercard, and EMVCo are all separately working on identity and trust frameworks for AI agents that make purchases on a user’s behalf, and building on top of that plumbing rather than trying to own it lowers Anthropic’s own liability while still giving it a foothold with major retailers ahead of the holiday shopping season.

The one number worth treating skeptically is Anthropic’s headline claim that retailers have seen shopping carts grow up to 35% larger and conversion rise 60% — the company has confirmed those figures came from a single partner rather than a broad, independently verified sample, so they’re better read as a best-case anecdote than an industry benchmark. Still, for developers, the practical value is real: a documented, forkable starting point for a genuinely hard integration problem is a meaningfully different offer than another chatbot demo.