Agentic commerce: When future customers are AI agents
Hanover-based start-up ShopAgentic aims to prepare fashion companies for the next stage of e-commerce development, known as agentic commerce. In the future, artificial intelligence (AI) agents will search for products, compare offers and prepare or even execute purchasing decisions. Consequently, brands must ensure their products can actually be found within these AI systems.
Alexander Ringsdorff and Kai-Thomas Krause, the two founders of ShopAgentic, bring extensive e-commerce and fashion experience to the table. They developed early mobile commerce solutions with CouchCommerce and later joined the founding team of the US omnichannel platform NewStore. With ShopAgentic, they are now tackling the next major upheaval in digital retail. The first client the company has publicly named is Munich-based luxury sportswear brand Bogner.
In this interview, Ringsdorff and Krause explain the new demands AI places on existing e-commerce shop systems, the panic moment they predict for the industry and how the balance of power between brands, retail platforms and consumers could shift. The concept of customer loyalty will also be entirely renegotiated in the AI era.
Mr Ringsdorff, let us start with you: why did you found ShopAgentic? Why is ShopAgentic needed in addition to the AI functions of established e-commerce platforms?
Ringsdorff: Established e-commerce platforms were built for a world in which humans shop via websites and apps. Agentic commerce changes this fundamentally. In the future, millions of queries will be generated by AI agents that automatically retrieve product information and trigger transactions directly.
This places completely different demands on infrastructure. ShopAgentic therefore initially complements existing systems and aims to provide various agents for multiple e-commerce use cases in the future. We process and enrich product data, providing information and interfaces for external AI search engines and shopping agents.
In the long term, we want to develop our platform into a complete standalone commerce system. Currently, however, we are focusing on making brands with their existing IT landscapes visible to AI systems, and eventually transaction-ready.
Why did you choose to specialise in the fashion industry? What makes this sector particularly interesting for the deployment of autonomous AI agents?
Krause: The fashion industry is particularly demanding because it is not enough to describe a product based on technical attributes. You also have to understand what a brand stands for, which target groups it appeals to and the significance of materials, fits and style aesthetics.
An example from our collaboration with Bogner illustrates this. There, product descriptions are sometimes created alongside photo shoots. If the actual product attributes change before the final collection is released, we can cross-reference visual assets and descriptions to identify discrepancies.
Our own experience also plays a role. We have supported numerous fashion brands over many years and therefore understand the specific nuances of the sector.
How urgent is the need for action for fashion retailers? Is it just fearmongering, or do retailers genuinely need to change their business models and technical infrastructures?
Ringsdorff: We are now convinced that this issue is substantial. A typical misconception is that companies only want to act once a significant proportion of their traffic comes from AI agents.
The problem, however, is that a brand that does not appear in the responses and recommendations of AI systems is not considered in the first place. Low agentic traffic can therefore actually be a sign that a brand is not yet sufficiently visible.
Krause: Generative Engine Optimisation [GEO, the optimisation for AI response systems; Ed.] therefore does not work via a few backlinks. Product data must be substantially improved and made understandable and accessible for AI systems. In addition, there are different platform requirements and protocols that retailers must continuously respond to.
I believe we will witness a real moment of panic when organic Google traffic noticeably declines due to AI search systems.
Which functions are already in use at ShopAgentic?
Krause: Our catalogue intelligence team, consisting of five AI agents, is currently in the beta phase with Bogner. Bogner is the first client we are permitted to speak about publicly. The agents take on different tasks. The Storyteller analyses brand identity, the Portraitist develops international buyer personas and the Merchandiser enriches the product catalogue with detailed information.
Another AI agent prepares the data for external shopping agents and protocols such as the Universal Commerce Protocol (UCP) from Google and the Agentic Commerce Protocol (ACP) from OpenAI. The fifth agent, our Mystery Shopper, analyses how visible a brand is within AI systems and under which terms it is recommended.
Together, these agents create a database that brands can use to improve their presence in AI systems. In the future, we will offer brands and retailers a comprehensive analysis showing how accessible and usable their online shops are for AI agents. We will examine whether AI agents can reach the shop, whether they can find and understand products and where improvements are needed.
What product information do AI agents need to recommend suitable fashion items, and what has your experience been so far?
Ringsdorff: AI agents require significantly more information than traditional product descriptions in online shops provide. Material, fit, occasion, style and brand positioning must be described precisely. Information about the buyers and their individual preferences is also required.
This is precisely what we are working on with ShopAgentic. We structure and enrich product data so that AI systems can not only compare products based on individual attributes, but also understand their context.
Krause: One initial concrete result from our collaboration with Bogner is automatically generated product FAQs, meaning answers to frequently asked questions. The brand now wants to use this information on its own website as well. This demonstrates that enriched product data offers added value not only for AI agents, but also for traditional customers.
Who will control the customer relationship in agentic commerce in the future – major platforms, brands or personal shopping agents?
Ringsdorff: We are currently witnessing a fundamental conflict over who will control the customer relationship in the future and in whose interest purchases are made.
Platforms like Amazon generate revenue not only from retail, but also from retail media. When personal AI agents such as Instinct or Meta's Muse research and purchase products without customers navigating the platform themselves, this advertising revenue comes under pressure. Amazon has therefore restricted external AI agents from accessing its platforms for the time being.
At the same time, a personal agent acts in the interest of its users, whereas a retailer's own agent inevitably pursues the interests of the platform operator. That is a conflict of interest. Whose agent am I actually using? The one working for me, or the one working for the retailer? I believe customers will be more inclined to trust a personal, platform-independent agent. However, the answer still lies in the future. In any case, I do not think they will want to use a separate assistant for every brand going forward. They will bring their own personal agent.
What does this mean for brands?
Ringsdorff: This also presents an opportunity for brands. If an AI agent knows that a product is available directly from the manufacturer and that a loyalty programme offers benefits there, it can consciously decide against a marketplace. However, brands must learn to cultivate their identity and customer relationships outside of their own websites.
Will AI agents intensify price competition in fashion retail?
Krause: In the future, agents will be able to compare prices much more transparently. However, this does not necessarily mean that the cheapest provider will always win. Availability, service, personal preferences and loyalty programmes also play a role. If an agent knows the purchasing behaviour and preferences of its users, it can factor these elements in as well.
A seller agent could, for instance, automatically make a personalised offer to particularly loyal customers. I even see this as an opportunity for brands to win customers back from marketplaces and better leverage their own strengths. Furthermore, there is a scenario that is actually advantageous for retail. If customers no longer come via platforms like Amazon, more margin remains with the brand. Personal agents could therefore actually deliver higher margins and reduce price pressure.
Where do limits of automation lie?
Ringsdorff: Particularly for brands, there are many tasks that humans should continue to handle in the future, such as creativity, curation, brand management and the development of new ideas. Essentially, an agent can analyse, repeat and combine what already exists. A brand, however, must continually reinvent itself.
Krause: I therefore do not believe the goal is to cut as many staff members as possible. Today, many highly qualified professionals in e-commerce spend a significant amount of their time on operational tasks such as data maintenance or recurring processes. If agents take over these tasks, teams can focus more heavily on value-adding activities. It is not about doing the same thing with fewer people, but rather achieving significantly more with the same team.
What technical hurdles currently remain in agentic commerce?
Krause: One challenge is that many agents today still shop like humans. They open websites in virtual browsers, analyse the interface and click through the shop step by step. This is terribly cumbersome, resource-intensive and slow.
Consequently, new standards are emerging, such as the Personal Agent Protocol (PAP), also known as Poppy. It is designed to regulate how personal agents are permitted to interact with brands, whether via traditional websites, direct Application Programming Interface (API) or Model Context Protocol (MCP) interfaces, or brand-owned agents.
This raises important security questions. Who is this agent, on whose behalf is it acting and is it authorised to access a customer account or execute transactions? These questions will significantly shape developments in the coming years.
What will a successful fashion retailer do differently in 2030 compared to today?
Ringsdorff: Our AI company brain came up with a rather good phrasing for this: a successful fashion retailer treats product data like a product in its own right. It is maintained, structured and machine-readable. This will become just as standard as having a responsive online shop is today.
Transaction capability is another factor. It will no longer be enough for a brand to simply be found in AI systems. It must also be able to allow agents to purchase from it directly.
Finally, the organisation itself will change. When AI agents operate in a trusted environment like our Agent OS, retailers will spend less time on operational, routine tasks such as data maintenance, allowing them to focus instead on curation, brand management and customer relationships.
OR CONTINUE WITH