Believe it or not, as of one year ago today, the term agentic artificial intelligence had never appeared on FinXTech.com. And then suddenly, it was everywhere. Since Sep. 30, 2025, the term has been mentioned in 20 articles or videos on this website. Coverage of agentic AI this year — and all the good or scary things it can do — has been all over the banking trade press and mainstream media. Banking groups have also been regularly discussing agentic AI at conferences and other events throughout 2026.
Yet most banks still aren’t using it.
While 72% of the bank CEOs, technology executives and board members surveyed for Bank Director’s 2026 Technology Survey reported that their institutions are using generative AI, just 30% say their banks are deploying agentic AI. Generative AI can create new content like text, programming code, images and audio based on user questions or prompts. Agentic AI is designed to allow autonomous agents to take action on their own to execute complex, multistep tasks with limited human involvement.
Jim Perry, senior strategist at Market Insights, a firm that provides data-driven consulting services to community banks and credit unions, says it is the autonomy that makes many financial institution executives reluctant to embrace agentic AI.
“We’ve had automated systems before, but when you start asking a system to act on your behalf, that changes the decisions,” he says. “Because then you’re talking about, ‘Well, what’s the agent going to do? What can’t it do? Who’s going to supervise it? Who’s going to be accountable when something goes wrong?’ So, I think there are a lot more calculations involved.”
All of the banks and credit unions FinXTech has interviewed about their use of agentic AI have said that humans are consistently reviewing the work performed by agents. Stu Bradley, senior vice president of risk, fraud and compliance solutions at the data and AI software company SAS, says that should be standard operating procedure. “We often get into the concept of, ‘Can agents really be used in an autonomous fashion? I think organizations need to think very carefully about getting to that end state,” he says.
Keeping Pace
Some early adopters of agentic AI say it has provided efficiencies for their lending teams. That is what $14.4 billion ConnectOne Bancorp, based in Englewood Cliffs, New Jersey, found as well. The bank has deployed an agentic AI tool from software provider nCino to handle tax return spreading, financial analysis, commercial relationship reviews and more.
While the most recent data from the Federal Deposit Insurance Corp. shows the average commercial bank runs at an efficiency ratio of 55%, ConnectOne Bank’s ratio currently sits at 43% — down from 49% just a year ago. (The lower the figure, the more efficient a bank is.) CEO Frank Sorrentino says agentic AI has helped improve the bank’s efficiency and believes it will continue to do so. “I don’t understand why people are questioning whether or not this is something we should be doing or not be doing,” he says. “And there’s only 30% [usage]? It should be 99%.”
While some banks have found early success with agentic AI, Bradley says many early adopters struggled to capture a meaningful return on investment. “They were thinking about technology first, and outcome and use case second,” he says. “And I think that’s why, over time, their [chief financial officers] were starting to ask questions like, ‘Where’s all my return on the investment that I’ve been making?’”
Bradley says the largest banks have since learned from their mistakes, and smaller financial institutions can as well. He advises them to outline specific use cases for agentic AI and the outcomes they want to achieve with it, before implementing the technology.
Governance and Data Are Key
A July International Data Corp. report commissioned by SAS found large international banks and other global organizations that built trustworthy governance practices for agentic AI tools were 15 times more likely to report a strong return on investment. Bradley says trustworthiness is based on what an organization can prove an AI agent is actually accomplishing.
“I would say there was a misperception in the marketplace that governance was actually stifling innovation,” he says. “But the reality is, if there was a consistent and established governance structure for analytics and AI at the enterprise level, and it was well thought through from an end-to-end standpoint, that actually it was a driver of innovation.”
Because an AI agent can only be as trustworthy as the data it is allowed to act on, banks and credit unions must ensure all their data is accurate, timely and accessible. If solid governance and data practices are in place, Bradley says a financial institution can then launch AI agents with the confidence they have a system in place to monitor and audit their work. “If you get those pieces right, then I do think you’ll have, whether you’re a large or small bank, the ability to scale your AI initiatives more effectively,” he says.
While that is the ideal scenario, it is not currently playing out that way for most banks. Of the 2026 Technology Survey respondents who indicated their banks are using agentic AI tools, only 48% are using custom tools built in-house; 83% said they are embedded in existing third-party applications. Perry says that can complicate matters. “They’re going to consume it through the core, through the CRM, through fraud tools, through other vendors,” he says. “That’s really going to require a whole other level of internal expertise just to understand what those systems are actually doing.”
That can also add to regulatory concerns. While there are no federal banking regulations on the use of AI, bankers and regulatory experts say examiners are asking banks to keep track of how they are using it and what risk frameworks they are putting around it. When it comes to AI provided by vendors or other third-parties, Robert Maddox, a partner at the law firm Bradley Arant Boult Cummings, advises clients to use the 2023 Interagency Guidance on Third-Party Relationships: Risk Management from the FDIC, Office of the Comptroller of the Currency and Federal Reserve as a guide. The agencies in September proposed an update on that guidance and are currently looking for comment.
There Is Plenty of Time
With the hype cycle surrounding agentic AI rapidly evolving, it might be easy for bankers who aren’t using it to feel as if they are falling behind. Chris Weidemann, chief AI officer and cofounder of Advisor Labs, says they shouldn’t. He works with small financial institutions to help them find simple AI solutions first, and says most are just scratching the surface with their generative AI tools.
“We look at agentic AI at kind of that 18-month point,” Weidemann says. “It’s like, ‘Once we get through low-hanging fruit, let’s start looking at the more complex things.’”
Perry thinks consumers will eventually force financial institutions to adopt the latest agentic AI tools in order to survive, but he believes banks and credit unions still have time to get it right.
“Even with the customer moving so rapidly on this, I still don’t think banks need to rush into [thinking], ‘We have to have this agentic workflow now. We have to figure this out, get it into production, make it happen,’” Perry says. “I think people have time to be really thoughtful about where and when they deploy this.”
Previous