Agentic Commerce and AI Shopping Agents – what changes in e-commerce (including B2B and Magento 2)

At the end of August 2026, Anthropic published an open source blueprint for AI agents in commerce. This is not another marketing note about “the future of AI”, but concrete, working code (github.com/anthropics/commerce-agents) that teams can use today. The repository ships ready examples for retail, commerce, telecom and entertainment.

For e-commerce companies – both B2C and B2B – this is one of the more important signals of recent months. The Agentic Commerce era is starting in earnest.

What is Agentic Commerce?

Classic e-commerce runs on one model: the customer searches → browses → compares → adds to cart → pays. Agentic Commerce flips that logic. The customer describes a need in natural language, and an AI agent:

• finds the right products,
• compares them,
• builds the cart,
• answers questions about availability, the returns policy or order status,
• and finally hands the customer over to the store’s existing checkout.

Anthropic describes two kinds of agent:

1. Shopping Agent – runs on the customer side (inside the store or app): it searches, compares, plans, assembles the cart and answers questions about orders and store policies.
2. Merchant Agent – supports the sales team: it explains sales performance, maintains product data, reacts to inventory and order alerts, and drafts pricing and campaign proposals for the team to approve.

The blueprint is available under the Apache 2.0 licence and works with Claude models in three ways: directly through the Messages API, through the Agent SDK, or as a Managed Agent connecting to your MCP server.

How does it affect the buying journey?

The biggest change touches three areas:

• Discovery – the customer does not need to know exact product names or filters. It is enough to say: “I need a tent, a sleeping bag and a camping stove for a weekend with two kids.”
• Product comparison – the agent does it on the fly, using attributes, descriptions and the relationships between products.
• Checkout – the agent assembles the cart, but payment and finalisation stay on the store’s side (which matters for security and control).

In practice this means the quality and structure of product data become a decisive success factor. An agent given incomplete, inconsistent or outdated information will perform poorly, or skip the product altogether.

AI Visibility – every store’s new obligation

More and more people talk about AI Visibility – how well products can be correctly understood and recommended by AI agents. If the same product has different descriptions, attributes or prices in the store, on marketplaces and in feeds, the agent gets contradictory signals. The result? The product drops out of recommendations.

This is exactly where PIM systems grow in importance. A well-implemented Product Information Management is no longer just a “content warehouse” – it becomes the source of truth used by both classic sales channels and AI.

What does it mean for B2B platforms?

In B2B the potential of Agentic Commerce is even greater than in B2C, but the requirements are higher too.

Complex catalogues, custom pricing, account hierarchies, order approval rules, units of measure, configurators and large recurring orders – all of this has to be understandable to the agent. Classic “nice product cards” are no longer enough.

B2B platforms (OroCommerce, Adobe Commerce B2B, Shopware B2B or bespoke solutions) will have to give agents not only product data but also business context: who can buy what, at what price, with which limits and under which approval procedure.

Companies that already have a tidy PIM and well-integrated systems will gain a clear advantage. Those keeping data in spreadsheets and many inconsistent sources will struggle.

What about Magento 2 / Adobe Commerce?

Magento 2 (Adobe Commerce) still powers a great many B2C and B2B stores in Poland and across Europe. In the context of Agentic Commerce, several points matter:

• Data quality in the Magento catalogue becomes critical – especially attributes, product relationships, variants and stock levels.
• Integration with a PIM (Akeneo, Pimcore, Ergonode) matters even more. Magento stops being the only place to manage product content.
• Magento’s GraphQL and API will be used more intensively by agents – it is worth keeping them performant and complete.
• Security (recall the recent StyleSmuggler) stays crucial – an AI agent must not become an extra attack vector.

Magento 2 stores that want to be ready for AI Shopping Agents should start now:

1. Put the product data model in order.
2. Implement or refine a PIM as the single source of truth.
3. Make sure the API returns complete, up-to-date information.
4. Prepare store policies (returns, delivery, warranties) in a form language models can read.

Which open source modules (ACP, UCP, MCP) already exist and where to start, we cover in a separate article: Why Magento 2 / Adobe Commerce handles Agentic Commerce well.

Summary

Agentic Commerce does not start with deploying an AI agent. It starts with the foundation – clean, structured, up-to-date product data and well-designed processes.

Anthropic’s open blueprint speeds this up and lowers the barrier to entry. Those who prepare their catalogue and integrations now will be able to test and deploy agents quickly. The rest will be catching up once competitors start genuinely gaining from larger carts and better conversion.

At Macopedia we have spent years helping companies build exactly this foundation: solid PIM systems, integrations with Magento 2 / Adobe Commerce, OroCommerce and other B2B platforms, and processes that keep data quality high.

If you want to check how ready your catalogue and architecture are for the AI Shopping Agents era – let’s talk.