Transcript
Hey everybody. Welcome back to another episode of What’s Going On in Banking. I’m Ron Shevlin, chief research officer at Cornerstone, and here, of course, with my co-host Stacey Bryant.
Stacey, how are you doing?
I’m freaking great.
I’m heading down to New Orleans for the New Jersey Bankers Association annual conference, and somehow a girl like me has never been to New Orleans before.
I’m ready to ask the ghosts on Bourbon Street what banking is going to look like over the next few months and, more importantly, where I should escape when all of this is over.
How are you doing, Ron?
Doing great.
And I’m a little bummed because I’m going somewhere new this week too: New Jersey.
Upper Montclair, to be exact.
That’s about 15 minutes from where you live, right?
Yes. I’m in Bergen County, so Montclair is close.
It’s beautiful, by the way. I’ll text you a few places you should try because the food is great.
What are you doing in my neck of the woods?
A bank is hosting an event for its commercial real estate developers and investor clients, and I’m presenting on what’s going on in commercial real estate.
It’s at the Upper Montclair Country Club, so I’ll be hobnobbing with the New Jersey elite.
Actually, it’ll mostly be bankers and their clients.
So we’re going to be two ships passing in the night. I’m bummed I’ll be in your neighborhood while you’re in New Orleans.
Anyway, let’s get into the banking news.
You were telling me about an article you saw on CNBC. Want to tee that one up?
Absolutely.
The headline was wild: “This bank CEO let his AI clone handle an earnings call. Now he’s signing an OpenAI deal.”
OpenAI has apparently found its first regional-bank test kitchen, and Customers Bank is handing over the keys.
The article says OpenAI engineers will work directly with the bank to modernize some of the slow, messy parts of lending, deposits, and payments.
That makes me think about the Industrial Revolution, except now the assembly line is AI-powered and the goal is to let bankers spend less time babysitting spreadsheets and more time actually banking.
If a roughly $26 billion institution is willing to let AI handle more of the back office so people can focus on relationships, why do so many boards still hear “AI” and think “cute chatbot on the website”?
Let’s dig into the article because the earnings-call part was definitely a stunt, but I do not mean that negatively.
Customers Bank is not a conventional, conservative bank.
Jay Sidhu founded and ran it for years. His daughter Lovleen eventually led the organization, and now Sam Sidhu is running it. It has always had a highly entrepreneurial culture.
What happened on the earnings call was interesting.
About half an hour into the call, Sam revealed that the prepared remarks up to that point had not actually been delivered by him. They had been delivered by an AI-generated clone of his voice.
I can imagine most bankers listening to this saying, “Yeah, that is absolutely not happening at my bank.”
Fine.
But the more important story is the OpenAI relationship.
A couple of things strike me.
First, I know a lot of IT leaders want to standardize on one AI platform. But I’m not convinced the industry is mature enough for most banks to make a huge long-term commitment to one model provider over another.
This is not Fiserv versus FIS. It is not Alkami versus Q2. It is not one mature loan-origination platform versus another.
These AI tools are changing too quickly, and the relative strengths of OpenAI, Anthropic, Google, and others are still evolving.
For Customers Bank, given its culture and appetite for experimentation, committing more deeply to OpenAI may be perfectly rational.
For a lot of other banks, I would move more cautiously.
I would love to sit down with Customers Bank’s risk, compliance, and governance teams.
The part that makes me nervous is how much data and process access the bank may ultimately expose to a third-party AI provider.
That raises very real questions around governance, model risk, compliance, security, and dependency.
Still, I give them credit for the stunt. Letting an AI clone deliver an investor call took guts.
If I find out that Steve Williams or Mazen at Cornerstone have been using AI clones on our Friday calls, I’m going to be upset.
Ready for the next topic?
Let’s keep going with AI agents.
In our last episode, we revisited several topics Jim Marous and I had discussed at the Financial Brand Forum, but we ran out of time before getting to agents.
You sent me a Substack article from some folks at Andreessen Horowitz that is worth discussing.
Tell us about it.
Yes. And first, another reminder about our monthly challenge. Ron and I are looking for the most thought-provoking or sharpest banking posts we see on LinkedIn, Substack, or elsewhere.
Tag us if you write something strong.
This article did not tag us, but it absolutely caught my attention.
The basic question was: as AI agents begin operating real systems, who is actually in control?
The article laid out several areas institutions need to think about.
The first is identity.
It is one thing to build a smart agent. It is another thing to know which agent you are interacting with, what it is authorized to do, and whether it should be trusted.
There is not yet a universal identity standard for agents.
The article discussed ideas such as Know Your Agent, or KYA, and using blockchain-style credentials or attestations to prove identity, permissions, limits, and track records.
It reminded me of when I was 17 and wanted to get into New York clubs.
For research purposes only, of course, I may or may not have gotten a fake ID from a mall in Jamaica, Queens.
It was basically my name typed onto a card and laminated.
I cannot confirm or deny whether it worked.
But that is the issue here. If agents are going to interact with each other, they need better proof than a laminated fake ID.
The second issue is governance.
If an agent is acting autonomously, what prevents the model provider or platform from changing something that alters the agent’s behavior?
Institutions need records of prompts, policies, training context, permissions, decisions, and changes.
They need ways to verify that the agent is still doing what the owner intended.
The third issue is payments.
Agents may eventually transact without a human interface. They could buy services, move funds, or interact with digital assets directly.
The article talked about stablecoins and on-chain settlement as possible infrastructure for those machine-to-machine payments.
And finally, there is verification.
If agents operate at a scale where humans cannot watch every move, how do institutions confirm that the outputs and actions are correct?
I thought the article was useful because it took agentic AI beyond the vague “AI agent” buzzword and started asking how identity, governance, payments, and control actually work.
What did you think?
I agree that it is a strong framework, and the authors are smart.
I also have a little built-in skepticism whenever venture-capital people publish something that perfectly aligns with the areas they are investing in. Sometimes the thesis and the portfolio are a little too conveniently aligned.
But the framework is valuable.
Banks will absolutely need agent identity, governance at the agent level, payment controls where relevant, verification, and user control.
I just think the article jumps ahead of where most banks are today.
Conceptually, a lot of this is not completely new.
Think back to the early days of e-commerce 30 years ago.
When I bought a book online, how did Amazon know it was really Ron Shevlin making the purchase? How did it know I was authorized to use the payment method?
We solved many versions of identity, authentication, payment authorization, and fraud over time.
Agentic commerce will require new versions of those controls, but we are not starting from zero.
And for banks specifically, the first wave of agents is probably not about agents independently going shopping or making payments.
It is more likely to be internal process work: account opening, fraud investigation, compliance, legal work, marketing operations, loan processing, servicing, and similar functions.
That means the immediate strategic problem is not, “How will our agent pay another agent?”
The more immediate questions are: where should we deploy agents, what work should they perform, and what business value are we trying to create?
Everyone asks whether agents will replace bank employees.
I think that is the wrong first question.
The better question is: how much faster can we redesign a process, at what cost, and with what operational risk?
The biggest benefit of agents may not be headcount reduction. It may be cycle-time compression.
A process that takes weeks today could potentially take hours or minutes.
Something that takes hours could take seconds.
That is a different source of value.
But changing a business process is not trivial.
Technology people sometimes act as though you can drop an agent into an organization and instantly change the way work gets done.
No. Processes involve people, responsibilities, controls, incentives, and change management.
Then there is cost.
Most banks already think they spend too much on technology.
Now we are asking them to pay for models, agent platforms, integration, orchestration, monitoring, security, and perhaps new vendors that do not sit natively inside their existing systems.
And then there is operational risk.
We all use ChatGPT and Claude, and both remind us that AI can make mistakes.
Agents may be operating on incomplete or imperfect data and making decisions more quickly than humans can review them.
So the issues in the article are real. I simply think the more immediate banking questions are pace of change, cost, process redesign, and operational risk.
I hear you, and I think your framing is useful.
But I also want to push back slightly on the idea that community institutions do not need to worry about the article yet.
The identity, governance, and verification issues reinforce why blockchain, tokenization, and related infrastructure at least belong in the strategic conversation.
I am not saying every bank needs to make a giant blockchain investment tomorrow.
But if these technologies may become part of how agents prove identity, permissions, ownership, or transactions, then executives should understand enough to include the topic in future-state planning.
Nubank and other digital institutions are already experimenting with blockchain-related infrastructure.
So even if the bank is not deploying those capabilities today, the article is another piece of the puzzle around what future banking architecture could look like.
That is fair, and I want to clarify my position.
I am not saying these issues are irrelevant to community banks and credit unions.
My concern is that a senior executive could read a piece like this and say, “Look at all these unsolved problems. This proves we should not touch agentic AI for another three years.”
That would be exactly the wrong takeaway.
The first question is still where the business opportunity is.
Which process could benefit? Which task could be improved? Which customer problem could be solved?
Then you build the controls around the use case.
And here is where I am going to quibble with your terminology.
No bank should make an investment in blockchain simply because it is blockchain.
Banks should invest in process improvement, cost reduction, risk reduction, customer value, and growth opportunities.
If blockchain is the best technology to support one of those goals, great. Use it.
But never start with, “We need to invest in blockchain.”
That reminds me of board meetings 15 years ago when someone would say, “My granddaughter is on social media. We need to be on social media.”
Wrong starting point.
Who are you trying to reach? What are you trying to accomplish? What problem are you solving?
Maybe social media is the right answer. Maybe it is not.
Blockchain is the same way.
Start with the business problem or opportunity, then choose the technology.
Okay. I think I was looking at the article from my Mother Teresa perspective. I want to save everybody.
And because I am from Brooklyn, let me borrow from Jay-Z too: I’ve got 99 problems, and having an agent should not automatically be one.
What I hear you saying is the same lesson we keep coming back to with AI more broadly.
Do not adopt something because everyone else is talking about it.
Start with what you are trying to solve.
I agree with that.
Great.
I think we can leave the main discussion there, but I am going to make one shameless plug before we close.
I started a Substack.
It is incredibly difficult to find. The address is simply ronchevlin.substack.com.
It is free. I am not putting up a paywall because I actually want people to read it.
I have republished a couple of FinTech Snark Tank pieces there to get things started, but there is original content too, and I plan to publish more.
I have very aggressive subscriber goals. I would like to get to at least 50 by the end of the year.
I think you can handle that.
Stacey, as always, I love doing this with you.
And to everybody listening, thank you for spending time with us. Please follow or subscribe wherever you listen, and we hope you’ll join us for the next episode.
If you enjoyed today’s episode of What’s Going On in Banking, follow us on Spotify, Apple Podcasts, YouTube, or wherever you listen.
We’ve got more witty, gritty conversations coming your way.
Stay tuned.
Enjoying What's Going On In Banking?
Subscribe on your favorite platform