- Key insights: Visa has partnered with OpenAI to allow for tokenized, agent-based transactions within its large language models; Mastercard pushed a new framework for machine payments; and Coinbase, like Robinhood, is opening its doors to agents to handle investments and payments.
- What's at stake: While banks take a cautious approach, the card networks are making it easier for consumers to stay within large language models. Meanwhile, fintechs are going agent agnostic in the hopes that their customers will bring their favorite agent onto their platform.
- Expert quote: "There is customer relationship disintermediation risk for the banks as AI agents become the primary banking interface, and customers may default to opening their LLM or their AI agent instead of their bank app to handle their transactions," Daniela Hawkins, a partner at Capco, told American Banker.
Banks have a lot of reasons to be
Agentic commerce is forging ahead,
This week brought a handful of new advancements for the burgeoning technology. Visa partnered with OpenAI to allow for tokenized, agent-based transactions within its large language models; Mastercard pushed a new framework for machine payments; and Coinbase,
Visa and Mastercard's plays
Visa's partnership with OpenAI injects the card giant's agentic commerce protocol, called Visa Intelligent Commerce, into OpenAI's LLMs to allow agents to make tokenized payments on its network to developers and merchants.
"Commerce is going to happen in many more places and in many more ways than it does today, and agents will play an increasingly important role in helping people complete tasks that involve money — from purchases and payments to more complex transactions," said Marco Mahrus, OpenAI's head of partnerships and commerce, in a statement.
Similarly, Mastercard, in conjunction with 30 other companies including Stripe, Global Payments, Solana, Checkout.com, and Adyen, released Agent Pay for Machines, a service that allows software to transact with software. For example, a small business owner may have its agent buy a domain, pay for hosting or creative assets, or set up checkout services within a budget.
Mastercard is also preparing for a new layer of commerce, where businesses create services for AI to buy and use.
"Agent Pay for Machines will create the conditions for a superbloom of AI business models," said Jorn Lambert, Mastercard's chief product officer, in a statement. "Machine payments can make it possible for services to be bought and sold among agents at fundamentally different scales than payments today — very high volumes, very small values, very fast and at extremely low latency."
For banks, the announcements are notable because the card networks are making it easier for consumers to stay within large language models.
"Visa is enabling the LLM before they're enabling their banks and the retailers." Richard Crone, CEO of Crone Consulting, told American Banker.
"They're pushing the payment 'off site' before they enable the agentic payment behind the firewall of a retailer or a bank," he said. "Off site" is a term used in retail to describe advertising and marketing efforts that happen outside of a retailer's own physical stores, websites, or mobile app.
That's good for the LLM providers who, like many platform-based technology companies, derive much of their value from the number of active users on the platform and their ability to monetize them.
"This is a big deal for ChatGPT, because if they can conclude the payment on their platform, then they own the entire right attribution loop. Nobody has to leave, [which is] super valuable to them," Crone said.
But it's bad for banks if they lose touch with their customers, Daniela Hawkins, a partner at Capco, told American Banker.
"There is customer relationship disintermediation risk for the banks as AI agents become the primary banking interface, and customers may default to opening their LLM or their AI agent instead of their bank app to handle their transactions," Hawkins said.
Consumers are already using agentic AI more, according to TD Bank's 2026 AI Insights Report. About 78% of Americans say they use AI in their daily lives, and 55% said they use AI to help manage their finances. That's up from 10% from just last year.
Fintechs get aggressive
At the same time, fintechs such as Robinhood and Coinbase are going agent agnostic in the hopes that their customers will bring their favorite agent onto their platform to trade and make payments.
"The way people interact with the internet is changing," Coinbase said in a release. "People are increasingly moving through the world via agents rather than apps — and businesses are rapidly rebuilding their products to be agent-first in turn. As that ecosystem grows, Coinbase for Agents is positioned to be your trading and spending account at the center of it."
Like Robinhood, Coinbase allows agents to manage investment portfolios, increase capital efficiency on cash positions and make data-informed trades. And that presents banks with deposit liquidity risk, Hawkins said.
"Quite frankly, the banks are underestimating their customers. They're underestimating their ability to use or move assets to some of these new capabilities," Hawkins said.
Crone believes the opportunity for banks and retailers is to bring the LLMs within their ecosystem in the same way that Robinhood and Coinbase has done.
"To survive, banks must get the large language models under their tent. Otherwise, ChatGPT Finances will eat their lunch because the one who enrolls is the one who controls," Crone said. "Banks already support open banking access for aggregators like Plaid. The next step is extending that infrastructure to govern inbound artificial intelligence agents."
There's also value in allowing customers optionality to choose their agent, Capco's Hawkins said, pointing to Microsoft's approach to agentic AI.
"Microsoft decided that Copilot calls all LLMs," Hawkins said. "Copilot can call Claude, it can call ChatGPT, it can call whatever you want, because they have a captured audience with Microsoft Office. Instead of trying to make Copilot compete with all these other models, [they] just have it be that interface."
Whatever banks decide to do, they'll need to do it quickly.
"If they don't make these updates to interact with these agents and these protocols through their systems, then they'll just become a ledger in the background. People won't even use the banks for anything anymore," Hawkins said.
