Financial institutions are preparing for a world in which AI agents interact with each other, creating significant risk and opportunity.  

The global market for agentic AI in financial services is projected to hit $80.9 billion by 2034 from $4.4 billion in 2026, according to research and consulting firm Market.us. 

From $1 trillion banks to community credit unions, FIs continue to roll out customer-facing AI agents. At the same time, banks and fintechs are increasingly deploying agents that act on behalf of consumers and businesses.  

Agent-to-agent communication is especially gaining traction in agentic commerce, Nina Owens, managing director of financial services strategy at Publicis Sapient, an AI and digital solutions company, told FinAi News.  

FIs are “extending their protections to an agent who is acting on behalf of a customer,” she said. “From the consumer side, it’s definitely moving that way.” 

For example, Citi and U.S. Bank have adopted Mastercard’s Agent Pay, which authorizes agents to make purchases on behalf of users. Similarly, banks such as HSBC, Santander and Commerce Bank are using Visa’s Agentic Ready program.  

Principle of trust 

Agent-to-agent interactions are inevitable because “the pace of change of tech is much faster than it ever has been during our lifetime,” Kaushik Gopal, executive vice president for business and market insights at Mastercard, told FinAi News. 

Consequently, the “principle of trust is so important,” Gopal said.  

“While there will be some discovery process in the financial ecosystem, it will result back to the first principles of, ‘I’m just going to work with the entities that I trust the most,’” he said. “As a trusted facilitator of payments, we are just trying to make sure, let’s set the standard. There are a lot of tech companies … so we’re trying to pull them in as well and educate them.” 

Emergence phenomenon  

Agent-to-agent communication is likely to create emergence phenomenon — complex, unpredictable behavioral patterns that arise from these interactions — highlighting the importance of specific guardrails, Andrew Bud, chief executive of iProov, told FinAi News.  

London-based iProov is an AI-driven identity verification platform that works with banks including ING, UBS, Rabobank and East West Bank.  

“If the guardrails on the agents that interact with the real world are appropriate, then the consequences are going to be contained,” he said. 

“Those guardrails themselves must be signed by genuine proofs of human presence because otherwise, agents will simply issue guardrails to each other.” 

— Andrew Bud, CEO, iProov

When agents issue guardrails to other agents without proper identity management, potential risks and compliance violations could spiral, Bud said.  

Identity management will become increasingly important if AI agents significantly outnumber employees at an organization, he said.  

“You’re going to need to move toward a paradigm that is fitted to the management of tens or hundreds of millions of agents,” he said.

“Inevitably, agents will interact with agents, and therefore the number of agents explodes, and the complexity of this environment explodes.

— Andrew Bud, CEO, iProov

“So, the management of agent identity will have to go from today’s federated models … to decentralized identity models,” Bud said.

Decentralized identity models refer to the idea of agents carrying cryptographic credentials that hold digitally signed proof of their identity, capabilities and authorization, he said.  

Margin compression  

One advantage of inter-agent communication is “margin compression,” meaning they will negotiate deals that are the fairest to each party, Brandon Arvanaghi, CEO of agentic AI-driven business banking platform Meow Technologies, told FinAi News.  

“Basically, the best deals possible are achieved,” he said. “You won’t have the ability to have a sneaky-high margin of service or product in the long run because agents will sniff that out. … The agent will know better, and they’ll delegate [negotiation tactics]. It’s like having a personal assistant on superpowers.” 

Handshake tradeoff 

Fintech Fenergo, an AI provider for compliance and onboarding solutions, is “absolutely thinking about agent interoperability,” Director of Market Development Garry Teekah told FinAi News. 

Fenergo works with banks including Citi, CIBC, U.S. Bank and SMBC Americas. 

Fenergo’s primary focus for agent-to-agent scenarios is to understand how to “validate that handshake tradeoff,” Teekah said.  

“Where does one agent start and the other one end?” he said. “Within that agent interoperability, the way that we’re thinking about it is, ‘How do you have that governance model that builds that system of record of validated data within that entity and the behavior profile of that entity?’” 

Those parameters would enable banks and customers to deploy agents that are best suited for specific scenarios, whether that’s data retrieval or account opening, Teekah said. 

“It’s important for banks to think about the actual use case and workflow,” he said. “Use the right tool for the job. Don’t use a hammer to nail in a screw.” 

Really more efficient? 

While agentic interactions could open numerous possibilities, banks must take a step back to determine if AI agents are more efficient, Will Rhoads, chief innovation officer at Brentwood, Tenn.-based Sonata Bank, told FinAi News.  

“How much of that communication is better handled via API?” he said. “Realistically, the consumers’ agent interacting with our API is probably a lot more efficient than us both burning tokens, asking each other questions back and forth.” 

Nonetheless, Rhoads said agent-to-agent communication will grow over time, especially as token costs decrease and agents become more API-driven.  

“We will soon enter a world where the [user interface] software that you use ceases to matter,” he said. “Instead, you’ll tell an AI agent what you want to do, and it gives you the [graphical user interface] you need to accomplish it.  

“At that point, that’s not just talking to each other. That’s everybody having bespoke software communicating with APIs across the world.” 

Register here for the FinAi Lending Summit, set for Oct. 7-8 in Las Vegas.