E-Commerce

Agentic Commerce: When AI Agents Shop in Your Shopware Store

Philipp HasseCEO & Lead DeveloperOctober 15, 20267 min read

Until now, online shopping worked like this: a person searches, compares, clicks and pays. With agentic commerce, an AI assistant takes over part of that. A customer tells ChatGPT or Gemini “Find me a warm winter jacket under €200 in size M”, the agent searches shops, compares offers – and in the next step, it puts the jacket into the cart itself.

Shopware shipped the first building blocks for this over the past year. At the same time, a lot is still in beta, and a lot depends on OpenAI, Google and others. In this article, we put into perspective what already works, what’s missing, and what you can do right now without jumping on the hype.

What exactly is agentic commerce?

An AI agent is an AI assistant that doesn’t just generate text but uses tools: it fetches product data, checks prices and availability and, if the shop allows it, can carry out actions. In one of its own articles, Shopware quotes Paul Krauss from the agency Team One with a fitting line: a model that only generates text doesn’t buy anything (Shopware: When AI shops for AI).

For an agent to “shop” in your store, it needs two things:

  1. Data for finding and comparing – usually a product feed with complete, machine-readable information.
  2. An interface for buying – a standardized protocol it can use to access the cart, checkout and order status.
Flow: a customer states a wish, an AI agent such as ChatGPT or Gemini searches and compares via the product feed and orders via a protocol like UCP in Shopware's agentic commerce sales channel. Feed available since Shopware 6.7.10, checkout via UCP as a beta extension.
From customer wish to order: the feed and the protocol are the two building blocks. Sources: Shopware Release News May 2026, Shopware Store.

The protocols: ACP and UCP

So that not every shop has to build its own interface for every AI provider, open standards are emerging. The two most important ones:

Agentic Commerce Protocol (ACP)Universal Commerce Protocol (UCP)
Initiated byOpenAI (with Stripe)Google (with Shopify, Walmart, Target and others)
IntroducedFall 2025, together with Instant Checkout in ChatGPTJanuary 11, 2026 at the NRF trade show
Where the agent buysChatGPTAI Mode in Google Search, Gemini
How shops connectProduct feed + checkout interfaceManifest at /.well-known/ucp, connection via API, Agent2Agent (A2A) or MCP

Sources: Shopware on ACP and Instant Checkout, InfoQ on UCP.

For merchants, this means you don’t have to pick “the one” protocol. Your shop software should speak the common standards, and that’s exactly where Shopware comes in.

What Shopware offers today

In spring 2025, Shopware co-founded the Agentic Commerce Alliance and is working with PayPal, among others, on agents that find, compare and buy products for customers (Shopware). Here’s what was actually shipped in 2026:

Building blockSinceWhat it doesStatus
Agentic commerce sales channelShopware 6.7.10 (May 2026)Dedicated sales channel with an OpenAI product feed (JSONL), a ready-made feed URL and tracking of orders that come in via AIin the core, more platforms announced
llms.txt and agents.mdShopware 6.7.12 (July 2026)Per-sales-channel files that explain to AI search engines and agents what your shop offersin the core
”Agentic Commerce (Beta)” extensionJune 2026, currently version 1.4.0UCP support for catalog, cart, discounts, checkout and orders; feeds for OpenAI and Googlebeta

Sources: Shopware Release News May, July, Shopware Store.

The sales channel from 6.7.10 automatically filters out invalid products and tracks traffic and orders from AI sources via the existing affiliate logic. That way, you’ll also be able to see later whether the channel is worth it at all.

What’s still missing – an honest look

As exciting as it sounds: for most shops, agentic commerce is still a testing ground in fall 2026.

  • The checkout is in beta. The “Agentic Commerce (Beta)” extension is explicitly labeled as beta. According to its store page, it requires at least one dynamic product group and access to packagist.org during installation and updates, and it can’t be installed in cluster setups. Several reviews reported errors after installation or updates. Shopware responded with fixes, and version 1.4.0 disables the extension when the SDK is missing instead of blocking the shop.
  • A feed alone doesn’t open a checkout. The feed makes sure agents can find and compare your products. They can only buy once a protocol like UCP is active as well.
  • The platforms have a say. According to Shopware, some of the AI providers’ features are currently limited to merchants in the US. Whether and when shops in Germany and elsewhere in Europe can fully participate is out of your hands.
  • Legal questions and liability are open. Who is liable if an agent orders the wrong thing? How do withdrawal rights and mandatory information work when no human sees the order page? There’s no established practice for this yet.

Why product data now matters more than your design

A person can be won over by beautiful images and a good home page. An agent sees none of that. It reads data, and it compares you in seconds with every other shop offering the same product.

For an agent to choose your product, it has to be able to answer the customer’s question with your data. “Waterproof?”, “Does M fit at 5’11”?”, “Delivery by Friday?” – if that isn’t clearly in the product data, your offer falls through the cracks.

Three stages to an agent-ready shop: stage 1 discoverable through feed, llms.txt and indexing; stage 2 comparable through complete product data such as dimensions, material, variants, GTIN, delivery time and price; stage 3 purchasable via a protocol like UCP, in beta as of October 2026.
You can reach the first two stages today, and they also help with Google and price comparison sites.

The good part: everything you do for agents pays off today. Clean product data improves your filters, your internal search, Google Shopping and how you show up on price comparison sites.

How to prepare your shop

Our checklist, sorted by effort, from “right away” to “once the channel is mature”:

StepWhat to doEffort
1. Keep Shopware up to dateAt least 6.7.12.1, ideally the latest version, so llms.txt, agents.md and the agentic commerce channel are availablelow
2. Check product dataRequired fields for every product: GTIN/EAN, manufacturer and manufacturer number, material, dimensions, weight, sizes and colors as real variantsmedium
3. Write copy that answers questionsDescriptions that answer concrete questions (care, fit, intended use) instead of pure marketing languagemedium
4. Maintain delivery times and stockAgents filter by “available by …” – outdated stock levels cost trustlow
5. Fill in llms.txt and agents.mdBriefly describe what you sell, to whom and on what terms (shipping, returns)low
6. Create the agentic commerce channelActivate the OpenAI feed and work through the error list of excluded productslow
7. Watch the trafficSee in the channel whether any visitors and orders are actually coming from AI sourcesongoing
8. Test checkout via UCPIn staging first, once the extension leaves beta and the platforms admit merchants in your countrylater

Conclusion: prepare, yes – hype, no

Agentic commerce is no longer a “someday” topic: the protocols are in place, the big AI providers are building in purchase features, and Shopware delivers a sales channel, feeds and a UCP connection. But for a typical mid-sized German or European shop, purchases by agents are still a thing of the future in fall 2026. The checkout is in beta, and the platforms are opening up step by step.

What you should do today is unspectacular, yet it pays off immediately: keep Shopware up to date, complete your product data, activate the feed and watch what happens. If you have your data under control now, you’ll be ready when the channel really starts generating revenue.

Want to clean up your product data, connect feeds or properly test the agentic commerce channel? Talk to us. We support you with custom Shopware development.

Sources

#Agentic Commerce#AI#Shopware 6.7#Product data#UCP