JPMorganChase: Agents will change the whole commerce stack

• Key insights: JPMorganChase is preparing for a world where AI agents fundamentally transform commerce. But to get there, the industry will need to agree on a set of universal frameworks and protocols. 
• What’s at stake: Agentic AI is slowly moving toward full autonomy, which is sparking questions around know-your-agent frameworks and interoperability as payments companies, merchant acquirers, card issuers and retailers work to create networks where agents can safely transact.
• Expert quote: “If you look across all the companies that are engaging in know-your-agent and the trust mechanism or intent conversations, everybody’s building for their slice of the pie… Our view is really that it needs to be universal.” — Zack Anderson, chief data and& analytics officer, global banking & payments, JPMorganChase. 

Agentic AI is advancing at a fast clip, and the country’s largest bank is preparing for a world where the budding technology fundamentally transforms retail and business-to-business commerce.

Processing Content

“We believe that the whole commerce stack will change over time,” Zack Anderson, JPMorganChase’s chief data and analytics officer, global banking and payments, told American Banker. “You’ll start to see things like know-your-agent [protocols] need to become a reality.”

Know your agent is a verification framework similar to know your customer requirements that identifies agents and verifies which person or organization is responsible for that agent, and which permissions and authorizations the agent has been granted.

Agentic AI is moving slowly toward full autonomy — some fintechs such as Robinhood, Coinbase and Meta’s Muse already allow consumers’ agents to make purchases and act autonomously on their behalf. And while actual agentic transaction volumes are still low, consumers are warming up to letting AI agents act autonomously, according to the Synchrony and Oxford Economics 2026 AI in Commerce study. About two-third of consumers surveyed said they would use AI if it came with more fraud protection, and 37% said they were already comfortable allowing AI to automatically purchase regularly used products.

“Consumers are open to AI taking on more of the shopping journey, but that willingness comes with clear expectations around security, transparency and control,” Margaux McLoughlin, research manager, thought leadership at Oxford Economics, said. “As AI moves from helping consumers make decisions to acting on their behalf, businesses have an opportunity to build trust into every interaction by giving consumers clear protections, visibility into how AI acts and meaningful control over the shopping experience.”

But in order for agentic transactions to truly scale, the industry will need to agree on a set of universal frameworks and protocols, Anderson said.

“If you look across all the companies that are engaging in know-your-agent and the trust mechanism or intent conversations, everybody’s building for their slice of the pie… Our view is really that it needs to be universal,” Anderson said. “We need protocols that can cross over, that don’t privilege either acquirers or card networks, that work in all the countries, and work for bank-to-bank and for the card networks.”

Universal frameworks are critical for widespread merchant adoption, Anderson said. “When you’re doing tens of millions of transactions as a merchant, you need to think about all the long tails that have big implications. You can’t cut off one payment method versus the other, one country versus the other, and you probably have multiple acquirers that are working in the background.”

The bank, as the largest credit card issuer and merchant acquirer in the country, is working to help the industry standardize some of these frameworks to benefit all parties involved, he said.

“In some ways, agentic commerce is throwing the cards up in the air, and they’re falling down in different places. And people are trying to influence how they fall down to privilege their position,” Anderson said. “That’s good from an innovation standpoint. But I’m not sure it’s good for either the merchants or the consumers.”

JPMorganChase , which was named the AI leader among banks for the fourth year in a row according to EvidentAI, has taken a slow and steady approach to agentic commerce. In March, it partnered with Paris-based software seller Mirakl that would allow its merchants’ product catalogs to be surfaced by AI.

The bank is also preparing for a world where AI agents reshape corporate treasury and business-to-business payments, an area where the technology is poised to advance much faster because there are less variables in the transaction.

“More often in that space, you know your counterparties, you know the accounts, you know a lot of the transactions are between banks as either soft or formal guarantors. There’s a lot more safety in that world,” Anderson said.

Trust will still be key to widespread adoption for B2B payments and treasury, though.

“The theory that we’re operating under is we’re assuming a high level of autonomy of agents working in treasury and cash management, and what we’re doing from the beginning is building in controls and escalations and mechanisms to enable that autonomy. Every treasurer will have to decide what their individual risk levels are,” Anderson said. “It’s pretty easy actually to create an agent that can move money. It’s hard to do it safely.”

JPMorganChase is developing a package of technical and policy controls that clients can select and use, he said.

“In reality, it’s not that different from the controls we have now for humans. Cash managers inside of a treasury operation have limits on how much they can send. They have maker checker rules. They have you no-release rules by a manager for the payment. They have limits on which counterparties and accounts they can have access to,” Anderson said.

“The trick is to take them out of that human realm and make them appropriate for agents,” he said.