As agentic commerce continues to advance, the payments infrastructure is working to keep pace. Wallet-based models are emerging as the most practical bridge between today’s payment systems and the autonomous purchases of the future. But that raises an important question: Who will own those wallets—AI companies or payments companies?
Christopher Miller, Lead Analyst of Emerging Payments at Javelin Strategy & Research, explores this question in the Will Agents Need Wallets? Why Payments Infrastructure Matters report. Miller recommends that payments incumbents focus on the value of the infrastructure layer rather than betting on a specific consumer-facing winner.
What Does It Mean to Be an Agent?
Many players in the agentic commerce space envision agents themselves—or at least the companies behind them—owning the entire customer relationship, from search and data management to transactions. The wallet is a distribution point for financial services, so whoever owns the consumer relationship there could also become the provider of their financial products.
Javelin’s research looks at the current state of agent adoption, how these services are being delivered, and what defines an agent provider. Providing the underlying models that power agents may not, in itself, represent a sustainable or differentiated business. Ultimately, the consumer-facing platform could matter more than the model behind it, with Apple Wallet, for example, potentially serving as the place where consumers access an agent rather than the agent itself becoming the wallet.
“If you’re a payments company and you sell credit and debit cards, you want to know the distribution points for your products and what it means for you if consumers begin to adopt something called an agent,” Miller said. “What company owns that customer and that customer relationship? That matters to merchants, it matters to payment processors, and it matters to financial product companies and the products they sell.”
The Model and the Brand
Many agents are already partnering with other services to incorporate wallet capabilities. Meta’s Muse, for instance, may use Stripe’s Link wallet to facilitate payments, but that relationship is largely invisible to consumers.
“If you are a company trying to convince someone to use Card X to make payments, you have to know where the stuff is stored so that you understand what the inflection points are, the decision points, where can you influence what card is being used,” said Miller. “Banks spend billions of dollars trying to buy their way to the top of wallet, to be the card that you think of, or even better, the one that you set as a default.”
Miller separates an agent’s functions into two distinct layers: the model and the brand. With a ChatGPT subscription, those layers are effectively one and the same, but that alignment isn’t essential. OpenAI could license its ChatGPT model to another consumer-facing platform. Perplexity draws on models from multiple companies.
Amazon similarly puts its own brand behind its agent, while the underlying model could come from Amazon, OpenAI, or Anthropic. Consumers may never know which model is powering the experience.
The distinction between the technology delivering the capabilities and the brand through which consumers access them is becoming more important. While those two layers have largely been aligned so far, that relationship is likely to evolve. As AI models become commoditized, the underlying technology is likely to become less of a differentiator. The consumer-facing brand and experience may carry greater weight.
“You won’t choose an agent because of the model,” Miller said. “You will choose one because it does some stuff better. And it may do some stuff better because they use a particular model or because it uses lots of different models in doing different things or whatever the case would be.”
Providing the Payment Capabilities
For payments companies, the strongest long-term opportunity lies in providing payment capabilities that work across agents, regardless of which platform ultimately wins consumer adoption.
Aligning exclusively with a leading AI company—by coming the wallet for ChatGPT, for example—creates significant exposure to that platform’s fortunes. If consumer preferences shift, the payments provider’s strategy could quickly lose its foundation. A more durable approach is to build infrastructure that can support transactions across multiple agents and platforms.
“This kind of stuff can change quickly,” Miller said. “Alta Vista was the number one search engine, then it went from the best to not really in existence in three years. The infrastructure layer is going to persist regardless of how these other things play out.”
The wallet itself could take several forms. An agent might operate its own wallet, rely on one provided by a third party or use a wallet connected directly to a merchant.
“One vision is you will have one agent that does everything for you,” said Miller. “I don’t think that’s guaranteed to be the way that people will actually consume agentic services. In particular, it may not be the way that they shop. Just because an agent could theoretically go to all the websites doesn’t mean that that’s what people are going to want. Maybe what they really want is an easier way to find what they want at Target.”
Retailers already have a head start in the latter category. Target, Starbucks, and Amazon have invested heavily in their own wallets, building products that incentivize customers to transact within their respective ecosystems. Those established payment relationships could serve as a foundation for agentic commerce, giving merchants a role in shaping how autonomous transactions are funded and completed.
“That’s why we separate the model layer from the brand layer,” said Miller. “Because if the model layer is just providing the capability, you can buy it directly from the people who make it. You can consume it through things that are provided to you by other people, such as Starbucks, Target, Walmart, whoever. Those are the open questions, and that’s why that separation is so important.”
