Agentic Commerce is emerging as one of the most significant trends in digital commerce. While AI has so far been used primarily to improve search, product recommendations, and customer service, autonomous AI agents could soon take over entire purchasing processes on behalf of customers. But what does that mean in practice for merchants, brands, and commerce platforms?
In this interview, Paul Krauss, Partner AI at Team One and an expert in AI-driven commerce, discusses the fundamentals of Agentic Commerce, the technological advancements driving its momentum, and the opportunities it creates for businesses. In the first part of our Expert Insights series, he explains why Agentic Commerce is gaining traction now and what changes merchants should already be preparing for.
1) What exactly is meant by Agentic Commerce – and how does it differ from classic ecommerce or Conversational Commerce?
Agentic Commerce shifts the abstraction layer in commerce. In classic ecommerce, we optimize the interface: better search, better filters, better product images. Conversational Commerce turned the interface into chat, but the logic remained the same: a human clicks through options, only in a text-based format. In Agentic Commerce, the machine acts on behalf of the customer; the human only formulates the intent.
I recommend Stefan Wenzel’s new book for anyone who wants to dive deeper into the topic, because many people think in either-or categories. Customer experience and efficiency always run in parallel, and Agentic Commerce is the most radical form of friction reduction that commerce has ever seen.
2) Why is Agentic Commerce becoming relevant right now? Which technological developments are making it market-ready?
Four things need to work at the same time for Agentic Commerce to move beyond demos: LLM capabilities, reliable protocols, payment, and of course user adoption.
Reasoning models have only become reliable enough in the past one to two years to be trusted with transactions that have real financial consequences – think of autonomous snack vending machines. Tool use, meaning the ability of a model to control external systems in a structured way, is the decisive lever here. A model that only generates text does not buy anything.
Then come the protocols. MCP from Anthropic defines how a model accesses external systems. ACP and UCP, developed by Stripe and OpenAI or Google, define the purchase process itself. In addition, there are x402 and the Machine Payments Protocol for payments between agents. Without these standards, every integration remains an individual project, and individual projects do not scale.
Third, payment infrastructure. That is exactly why the wave is starting with payment providers, not with merchants. Payment builds the rails for transaction traffic. Providers are using either virtual credit cards and/or shared payment tokens, which usually work independently of payment method and processor, with clearly defined limits. This is the basis for allowing an agent to pay in a trustworthy way.
Stablecoins make microtransactions between agents practical for the first time, and that is more than a technical detail. It opens up a type of business that did not exist before – and that may only fail because of adoption. But the major tech platforms are going all in. There will not be a cautious rollout. Agentic Commerce with advertising models could at least partially amortize the high training costs of frontier models.
3) What specific role do AI agents play in the purchase process – do they only advise, or do they actually make independent decisions?
Both, depending on the level and the product category. Today, most agents operate at levels one to three: filling out forms, searching semantically, remembering preferences. At its core, this is better autofill plus a smarter recommender. Real delegation from level four onward is something we are seeing only in isolated cases, for example when ChatGPT Shopping completes an Etsy purchase via ACP or when Perplexity sells directly in chat using PayPal Instant Buy.
Levels four and five were the scenario I outlined last year in my K5 theses.
Realistically, we are currently between levels two and three, and even then only in certain product categories. For consumables such as printer cartridges or dog food, delegation is closer. For clothing or furniture, it is still far away. Adoption is very uneven, and anyone speaking broadly about Agentic Commerce today is mixing together five levels that are, in reality, five different markets.
This needs to be broken down strongly by market segment. For commodity products – batteries, printer paper, consumables – full autonomy will become the norm. There is no emotional value in the decision. People do not want to be involved. The agent optimizes price, delivery time, and availability. In lifestyle and luxury, the model will be hybrid: the agent creates the shortlist, the human decides. In B2B – and this is underestimated – the agent has the greatest potential, because specifications and contract logic are already structured. That drastically reduces the risk, and many liability risks can be handled deterministically through rules.
Data also shows that 93 percent of consumers still prefer human interaction for complex problems. Human-in-the-loop will become a luxury good – something anyone notices when they try to speak to an actual person in customer service. Standard transactions will increasingly move to agents. Premium offerings will differentiate themselves through real, personalized, good, and/or human service.
Replenishment is the obvious use case, but the truly interesting one goes further. The CEX Trendradar 2026 describes, among other things, a banking app use case in which an agent aggregates a customer’s subscription landscape, cancels unused subscriptions for the user at the click of a button, and, if requested, routes the freed-up amount into funds or savings products. That is an agent killer use case, because the customer is not “buying something”; they are reallocating liquidity. The agent acts on the spending side and the investment side at the same time. Whoever offers this early builds a relationship that no traditional merchant has ever had – not through the product, but through the money behind it.
4) How does Agentic Commerce change the relationship between brand, platform, and customer when AI increasingly acts as an intermediary?
The customer funnel was never a law of nature, and the interface is disappearing while the decision point shifts downstream. Those who used to do SEO are, at best, only half-visible in AI Overviews. Affiliate, couponing, price comparison – all these intermediary layers are breaking away because the agent performs their function internally. What used to be laboriously “guided” is now answered directly.
The lesson from the ECC figures is important. 43 percent of German consumers have heard of AI shopping agents, and 61 percent can imagine using them. But only 11 percent say “definitely,” while 50 percent say “for certain purchases.” That means consumers are already thinking by category. They do not want to delegate their whole life. They want to delegate selectively.
Platforms remain the gatekeepers. In fact, they become even stronger. What used to be the search results page is now the answer from an agent. There are no longer ten blue links. Organic reach is steadily declining.
For brands, this means a double shift. Externally, they are no longer fighting for shelf space or advertising placements, but to be anchored as superior in the model’s latent space. When the agent asks, “Should I choose brand A or B?”, brand A must already be encoded in the model as the right answer, or there must be a clear customer preference as part of the prompt or clearly visible in the purchase history. That is a different discipline from classic performance marketing.
Internally, merchants are not losing reach, but relevance in the decision-making process. From the agent’s perspective, most products are functionally substitutable. If done well, the agent knows the customer better than the merchant does – not through clickstreams, but through situational understanding and persistent context. Who owns these agents? That question is becoming strategically more important than any conversion optimization of the past ten years.
5) What opportunities do you see for merchants and manufacturers – for example in conversion, personalization, efficiency, or customer loyalty?
There are two viable paths, and as always, the danger lies in the middle. Either relevance: the brand is so clearly positioned that the customer explicitly tells the agent, “only buy there.” Or efficiency: you are the easiest-to-read warehouse in the world, with perfect data, clean APIs, and fast logistics. Everything in between will be sorted out by the algorithm and the customer alike.
In conversion, Agentic Commerce removes a large part of the friction that causes revenue to be lost today. No more cart abandonment, because there is no cart. No address forms, no checkout drop-off. Stripe already demonstrates this measurably with Stripe Checkout and Shopify Checkout, whose success is based on the familiarity effect and higher conversion rates.
In personalization, replenishment becomes the killer application. There is a very large class of products where consumers will accept it if AI suggests: “Buy now, it is cheaper” or “you are about to run out.” That is genuine added value, not forced convenience.
In my view, B2B has the greatest underestimated potential. Procurement is highly structured, contract logic can be standardized, and specifications are available in machine-readable form. This is not about emotion, but about process costs – and that is exactly where agents scale best. Anyone using agents today in manufacturer-to-merchant relationships gains efficiency advantages that will only arrive in the consumer market years from now.
Differentiation also moves downstream, to where it is truly felt. Logistics becomes a brand moment, service becomes a value center, and communication becomes a relationship rather than an announcement. Exactly where many merchants have tried to save costs in the past is where loyalty will be decided in the future.
Agentic Commerce promises faster, more efficient, and increasingly automated purchasing processes. At the same time, it is fundamentally reshaping the roles of merchants, brands, and commerce platforms. But with these new opportunities also come new challenges.
In the second part of this interview, coming in two weeks, Paul Krauss explores the risks of autonomous purchasing decisions, explains how businesses can prepare for Agentic Commerce, and discusses the regulatory questions that are likely to shape the future of AI-driven commerce.
