Anthropic has open-sourced Claude Commerce Agents, a set of reference code that merchants can use to build their own shopping and operations agents. The repo ships with Apache 2.0 licensing and no direct payment-collection interface. A consumer-facing agent handles product search, comparison, bundling and adding items to a cart; a merchant-facing agent covers sales, inventory, pricing and marketing tasks. The shopping agent can only hand the cart to a merchant's own checkout page. Code-level rules also gate payments, product sources, price-change ranges and human approvals, rather than relying on prompt instructions alone. The merchant agent can generate proposed price changes, restocking actions or campaign edits, but those changes land only after manual approval. Anthropic advises against splitting search, returns or pricing into many sub-agents; instead, one Claude holds the full conversation and loads different Skills as needed. Anthropic says comparisons across enterprise deployments show this design delivers higher quality, generally lower token consumption and lower latency. One partner reported shopping-cart sizes up 30–35% and customers' purchase-completion probability up about 60%.
Anthropic has open-sourced Claude Commerce Agents, offering merchants a starting point for building shopping and operations agents of their own. The reference implementation is released under Apache 2.0, and it deliberately leaves out any direct payment-collection code.
Two agents, two workflows
One agent in the set points at consumers. It can search products, compare options, put together multi-item combinations and add them to a cart. Another agent faces the merchant, covering sales, inventory, pricing and marketing. The shopping agent's only handoff is the merchant's checkout page; no payment interface is included in the repo.
Those guardrails are not just written into prompts. Code-level checks cover payments, product provenance, price ranges and human approval. A merchant agent wanting to change prices, restock items or update campaigns can generate pending changes, but nothing takes effect until a person approves it.
One Claude, loaded Skills
Anthropic says it does not recommend breaking search, returns or pricing into many small sub-agents. Instead, a single Claude keeps the full conversation context and pulls in different Skills only as the task needs them. Across several enterprise deployments, Anthropic says, this setup has produced higher quality output while generally using fewer tokens and showing lower latency.
The company adds that one partner, after going live, saw shopping-cart sizes rise roughly 30–35% and the probability that a customer completes a purchase climb about 60%.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
