Elastic Path is the API-first, AI-ready commerce platform for brands with complex requirements: large catalogs, multi-tier pricing, and multiple brands and markets. Manufacturers, distributors, and retailers run account hierarchies, quoting, and negotiated pricing natively, not through bolt-on apps and workarounds.

Last week we hosted Agentic Island, a virtual summit backed by Shopware. Over two hours I heard from Shopware co-founder and Chief Agentic Officer, Stefan Hamann, and four agency and merchant pairs about what agentic commerce looks like inside real projects. Three things stuck with me.

An agent can’t recommend or buy a product it never finds. Stefan pointed out that an agent usually puts about three options in front of a shopper, so the competition to be one of them is tight.

His suggestion was to ask several people in your company for every reason a customer should not buy a particular product. Most merchants have never written that down. An agent matching a product to a situation needs to know what the product is bad at, and most product pages only cover the good.

A cheap first step is to ask ChatGPT and Claude about your own category, the way a customer would, and see whether you come up. After that, run the free readiness scanner and fix the technical basics it flags.

Agentic Commerce Alliance: https://www.agentic-commerce.org/
Agentic Readiness Scanner: https://agenticcommerce.shop
Agentic Commerce Maturity Index: https://acmi.agentic-commerce.org/

Four case studies were presented, and only one was close to agentic. The Mall Street, a new premium marketplace in New York, has an AI layer called Pathfinder that builds a mood board from images a shopper uploads. It was built from scratch, with no legacy data to move.

The other three were mostly about replatforming. PartsTree moved more than 4M parts, 450,000 models and data from more than 150 OEMs off a custom platform it had run for 25 years. ElleVet Sciences brought three sites onto one platform and moved its pet supplement subscriptions over, and partway through, its payment provider dropped it. SANY America told us plainly that its dealer catalog project has no AI in it yet.

That seems like the right order to me. If your catalog lives in a 25-year-old document database, an agent is the last thing you build. At PartsTree and ElleVet, the data migration was the hardest part of the project, and the agentic work they want to do next depends on it.

At Retail Club’s AI Festival in Huntington Beach, I asked a room of more than 30 retailers and brands what share of commerce would be agentic by the end of 2027. Every one of them said under 5%. Stefan’s view is that roughly 30% of buying will be in the hands of agents by 2030, with B2B moving faster than consumer retail.

I think the room was a little low. At the start of this year, few of the companies I work with would have predicted how much they use AI today. Most of us struggle to picture an exponential curve, and I think this is one. Stefan also expects agents to be buying on people’s behalf within months. His 30% still depends on merchants finishing the foundation work from lesson two.