Most banking executives thinking about AI asked the same question: which vendor should we partner with? Booshan Rengachari asked something different: where in the payment stack does AI actually belong?

Rengachari founded Finzly in 2018. The company builds payment infrastructure for banks still running on systems from the 1980s. By 2023, he kept seeing the same story play out. Banks would bolt intelligence onto existing payment operations, run pilots, and get promising results. Then they hit a wall. The AI could not access the real-time transaction data where decisions actually happen. Most payment processors responded with AI dashboards and after-the-fact analytics. Rengachari saw an architecture problem.

In October 2025, Finzly launched Agentic Galaxy. These AI agents operate inside the payment flow itself, not as a layer on top. In early deployments, operations teams report moving from hours spent investigating flagged transactions to resolving exceptions in seconds, with humans retaining control for compliance.

In this conversation, Rengachari explains why embedded AI architecture beats add-on tools, why banks need AI that operates rather than assists, and what his bet on stablecoins and programmable money reveals about where payment infrastructure is actually headed.

Foundation: Understanding Finzly and the modernization of a broken system

1. For readers who may not be familiar with Finzly, can you start by explaining what problem you’re solving for banks and credit unions? What does modern payment infrastructure mean in practical terms?

The banks are running payments on technology that was built for a world of analog communication and business hours. ACH settles overnight and wires need manual review. 

If you’re a bank with aspirations to provide instant payments, good luck if your mainframe goes down at 2 a.m., which it does at most banks.

The reality is, banking is changing faster than ever – with blockchain, tokenized assets, real-time payments, and new ways to deliver services like embedded banking. The problem is, most banks react to every new trend by bolting on point solutions or adding to their legacy systems. That just creates complexity, slows innovation, and makes it harder to scale.

The smarter approach is to build one solid innovation core. That’s why we built BankOS, to help banks  Imagine running ACH, Fedwire, SWIFT, RTP, FedNow — and tokenized assets — all from the same platform with a single API. Suddenly, you’re not juggling five or 9 different processors for five or nine different rails. You’re running one system, with full control and visibility.

2. What’s actually broken in traditional payment systems that makes modernization so urgent?

People assume because their banking app works, the plumbing works. It doesn’t and certainly not in the way it needs to for a world of instantaneous payments. 

Many banks run legacy cores, often on mainframe technology, decades old. As banks have needed to become more digital they have bolted on different tech infrastructure and solutions creating a highly complex maze of legacy technology. Unfortunately, these cores became so vital to the running of the bank, yet so complex, that switching to a new core is the equivalent of conducting open heart surgery while the patient is still awake. 

Our “Surround and Shrink” approach lets banks run Finzly alongside their legacy systems. Migrate one rail at a time. No big bang cutover. A recent webinar we ran showed 70% of participants voting they not confident their infrastructure can handle modern payments. 84% said it’s time to change. After decades of stagnation there is movement happening. 

3. What do BankOS and the set of Galaxies it powers offer, and how do they differ from what banks have been using?

BankOS is the underlying operating system, the innovation core that isthe foundation for how a bank moves money. . From BankOS, the Galaxies like Payment Galaxy, Account Galaxy, and Digital Galaxy provide modular, easy-to-launch building blocks that let banks deploy new capabilities quickly.  

The difference from what banks have today is pretty stark. Most institutions run a separate processor for ACH, another for wires, another for instant payments, maybe another for international, and now one more for tokenized money movement. Each one has its own interface, its own compliance workflow, its own headaches. We collapse all of that into one platform with one API. And we’ve validated with AWS that it handles volumes comparable to the Big 4 U.S. banks, so it’s not a small-bank-only play.

With BankOS and its Galaxies, banks can offer a full spectrum of capabilities from a single modern operating system, manage both traditional and tokenized assets seamlessly, and continuously adapt to new opportunities and regulatory requirements — all without layering on legacy complexity. It’s a foundation for both day-to-day operations and long-term innovation.

The AI Inflection Point

4. What does embedded AI actually look like in practice, and what makes these systems “agents” rather than just sophisticated automation?

Most of what passes for AI in banking right now is analytics after the fact. A dashboard that tells you what happened. That’s useful, but it doesn’t change how payments actually move.

What we did with Agentic Galaxy is put the intelligence inside the payment flow itself, in the transaction lifecycle and the bank operations that manage these flows. 

The “agentic” part matters because traditional automation is rigid. If X, then Y. An agent looks at context, weighs multiple variables, and acts or recommends an action based on what it’s learned. The critical piece is we keep humans in the loop for compliance. The agent doesn’t operate in the dark. Gilles Ubaghs at Datos Insights described it well — it helps banks move from reactive positioning to something more like continuous evolution.

5. Where in a typical payment workflow do your AI agents operate, and what decisions are they making?

When a $500,000 payment is sent, a series of decisions must occur: rail choice, ISO 20022 formatting, compliance, and beneficiary bank capability.

Legacy systems require manual handling for each step. Operations staff manually resolve formatting issues (looking up codes) and spend hours investigating compliance flags before a payment moves.

With our AI agents, much of this work is automated, but humans remain in the loop for true exceptions, receiving full context so they can act quickly and effectively. The result is faster, smoother, and more accurate payment operations.

How AI Performs in Production

6. Can you share a specific example where AI delivered significant time savings?

Where we see the biggest impact is exception handling. Picture a flagged wire sitting in a queue. Someone has to pull up multiple screens, cross-reference data, figure out what’s actually wrong. That can eat hours. Our AI-powered workflow and Payment Ops capabilities automatically handle routine tasks like routing payments, generating reports, and flagging exceptions. This significantly reduces manual effort — operations staff only step in when there’s a true exception, allowing banks to accelerate wire processing times by more than 10x.

7. How do your AI agents learn to resolve exceptions?

 Our AI agents learn from past exceptions—tracking how they were resolved and using that feedback to improve future decisions. This helps reduce false positives and increase the straight-through processing rate over time.8. What about quality metrics — fewer failed transactions, lower exception rates?

I can point to what we’ve already achieved without AI — 100% straight-through processing on Fedwire. Zero manual intervention on wires. That’s a bar most legacy processors aren’t coming close to clearing. As AI agents get more production time, we expect those numbers to improve across exception rates and fraud detection. It’s too early to provide concrete performance stats right now but we’re getting close. 

9. What are the architectural decisions competitors can’t replicate quickly?

Our AI sees across all rails—ACH, Fedwire, RTP, FedNow, and SWIFT—from a single platform, giving us visibility competitors simply can’t replicate. On top of that, we built the system on ISO 20022 from the ground up, not layered onto legacy formats. Real-time, 24/7 payments data feeds our AI, so its intelligence is deep, accurate, and immediate—something you can’t achieve with flat files or older systems.

10. How do you determine what AI handles autonomously versus what requires human judgment?

Data enrichment, format conversion, smart routing, AI can handle those on its own. The risk profile is well understood and the criteria are clear.

Most exception handling sits in the middle right now. The AI investigates, pulls context, and recommends an action. A person reviews and confirms or overrides. That’s where the biggest time savings come from.

Then there are things that stay fully human such as regulatory interpretations, unusual escalations, policy calls where the nuance matters too much to automate. That boundary will shift over time as confidence builds, but we’re not going to remove humans from compliance-sensitive decisions. Banking runs on trust and accountability.

Security, Privacy, and What Could Go Wrong

12. If an AI agent makes a decision that results in an error, how do you handle accountability?

Every action gets logged: what happened, what data informed it, what triggered it, whether a human reviewed it. This creates a full audit trail. Everything on our platform is explainable and auditable.

Stablecoins and Programmable Money

13. What’s driving the move into stablecoins?

There’s just a solid business case argument, frankly. What stablecoins offer that traditional rails can’t is dramatically cheaper and faster cross-border transfers. Programmability is also key, this involves payments that trigger automatically based on conditions, which traditional rails just can’t do. 

14. How do AI agents interact with programmable money capabilities?

They’re a natural pair. Stablecoins give you rails for conditional, automated payments. AI gives you the intelligence to determine when and how those conditions should fire.

Think about a supply chain payment. Shipment arrives, gets verified, stablecoin payment releases automatically per a smart contract. Now add an AI agent that can assess whether the shipment data matches historical patterns, flag anything unusual, optimize timing for FX rates, and check compliance across jurisdictions. All happening in real time.

Adoption and Looking Ahead

15. Can you share early adoption signals from Agentic Galaxy’s first four months?

What I can tell you is the nature of the conversation has changed. A year ago, banks were asking whether they should use AI. Now it’s about deployment timelines. And they want something purpose-built for payments, not a generic AI tool someone’s trying to wedge into banking.

16. How does AI transform the BankOS vision over the next few years?

We’re building toward a banking OS that actually learns. 

Near term, that means AI agents absorbing the bulk of routine operations such as exception handling, compliance pre-screening, routing, reconciliation. The operations team shifts from doing the work to overseeing the intelligence that does the work.

Further out, I think about things like predictive liquidity management, where the system anticipates funding needs before they arise. Combine that with stablecoins and tokenized deposits, and banks can offer service categories that genuinely don’t exist yet. Automated treasury management. Conditional cross-border settlements in real time.

Origianl Creator: Genaro Palma
Original Link: https://justainews.com/industries/finance-and-banking/interview-booshan-rengachari-founder-and-ceo-of-finzly/
Originally Posted: Sat, 28 Feb 2026 11:37:42 +0000