Agentic AI changes who clicks the payment button
AI assistants such as ChatGPT, Gemini and Copilot are just starting to research, select and pay for products on behalf of their users. For merchants and payment providers, this means that for the first time in thirty years, the checkout has to deal with someone who is neither human nor a classic interface. In this article, we explain what this means in concrete terms – and why it is better for merchants to deal with it now than in two years’ time.
What it's all about: A simple example that explains a lot
Imagine someone saying to ChatGPT: “Find me the cheapest running shoes in size 44 that are with me until Thursday and buy them on account.” Today, the convenience ends where the AI returns a list of links and the user has to click on them themselves. Tomorrow, the AI will process everything up to the paid shopping cart – without the user having to take any further action.
For this to work, three things have to fit together:
- The AI must find the right products.
- It must be able to trigger a payment without knowing card details.
- The retailer must recognize that a machine is buying – so that they know how to deal with it.
New technical standards have emerged over the last twelve months for precisely these three things. Stripe and OpenAI have published the Agentic Commerce Protocol (ACP), Google the Universal Commerce Protocol (UCP), Visa has launched Intelligent Commerce and Mastercard Agent Pay.
What sounds like technical details has a very practical consequence: merchants who want to be found, selected and supplied by AI assistants have to adapt their own infrastructure. Those who do not do this will simply not appear in the recommendations because the machine cannot read or address the checkout.
Who or what decides whether AI purchases work safely?
Discovery and product data are one thing. But as soon as money starts flowing, questions come into play that have grown over the years in traditional online retail. And these need to be answered in a completely new way for AI purchases.
- Who actually presses the “Buy” button?
In a traditional checkout, a person enters their card details. With an AI agent, things are different: the card belongs to a human, but the click is made by software. In order for retailers and banks to still be allowed to approve the payment process, they need new ways to recognize the AI as a legitimate representative of the customer – without showing it card data.
- Who does the detective work?
Today’s fraud detection relies heavily on human behavior: Typing speed, mouse movement, device fingerprint. All of this is eliminated when an agent makes a purchase. At the same time, new risks open up, such as someone using a compromised agent to make purchases on behalf of the victim. The entire risk setup has to be restructured.
- Who is liable if something goes wrong?
If the agent orders the wrong shoe size, books the wrong date or reserves the wrong room at the hotel – who bears the costs? The customer, the retailer, the agent’s provider? These questions have not yet been clarified in regulatory terms. Those who have clean processes early on will have significantly less friction later on.
Indispensable for Agentic Commerce: Clean product data
When a human visits a store, they can improvise with fuzzy texts, unusual layouts and missing information. An AI can’t do that. It reads structured data – i.e. clearly defined fields with unambiguous content. If the delivery date is missing as a data field, the product simply won’t appear in queries such as “in my store by Thursday”. If the return policy is missing as structured information, the agent cannot sort the product into a recommendation “with free returns”.
This shifts something fundamental: In traditional e-commerce, visibility was a matter of SEO, marketing budget and conversion optimization. In a world where an agent compares tens of thousands of products, visibility becomes a question of data quality. This applies to price, availability, shipping, dimensions, materials, return conditions – everything that previously appeared as “free-flowing text” somewhere on the product page.
In concrete terms, this means that a shoe retailer with perfectly maintained product feeds is preferred by the agent over a competitor whose data is incomplete or poorly structured – even if their products are objectively better or cheaper. The agent does not choose the best product. He chooses the one he understands.
Agentic AI × Digital Payment: Where the market stands today
The situation is confusing. And that is part of the problem.
A single standard has not prevailed. Instead, various players are developing their own protocols in parallel. For retailers, this means that betting on a single horse is risky. Supporting several in parallel is costly.
Four current developments in agent-based commerce that are having a lasting impact on digital payments:
Conclusion: Agentic Commerce is not waiting for clarity
Nobody knows which protocol will prevail. This is precisely why a clear strategy is needed.
At such an early stage, the biggest danger is not doing the wrong thing. It is to do nothing at all because you are waiting for clarity. This clarity will not come in the next two years.
What will come are the first competitors whose products suddenly appear more frequently in AI recommendations. The first banks that specifically issue agent-enabled cards. The first complaints when an agent orders the wrong thing and nobody knows who is liable. The companies that take a pragmatic position now – not solving everything at once, but building the architecture in such a way that it can react – will build up a measurable lead in this phase.
What to do now: Equip payment services for Agentic Commerce
valantic’s digital payments experts provide support in all steps that are necessary for secure payment processes in the age of agent-based commerce: from assessing whether it is worthwhile for you to become active now – and if so, where – to implementing innovative payment services.
Three steps for merchants to become Agentic-AI-ready:
30 minutes is enough to find a sensible direction. Together, we look at your setup, your sales channels and the protocols that are relevant to you. Then we’ll get down to implementation. Are you going with us?
Turn your vision into reality: Female Digital Impact
