Do Banks Need AI Employees More Than They Do Core Upgrades?

TL;DR

Buying for a business?Offer from Amazon

Get business pricing on office and shipping supplies

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

Kastle CEO Rishi Choudhary says banks can use AI agents to carry out work across existing core systems rather than wait for costly infrastructure replacements. His proposal, discussed in a PYMNTS interview after Kastle’s $24 million Series A raise, depends on restricted permissions, reliable performance and human oversight; the interview does not establish that agents can replace core upgrades or deliver proven savings.

Kastle CEO Rishi Choudhary says banks can deploy AI agents to perform work across existing core banking systems instead of making major core replacements the first step in modernization. The proposal, described in a PYMNTS interview following Kastle’s $24 million Series A raise, frames agents as a new operational layer—not a substitute for the systems that hold bank data and business logic.

Choudhary described agents as a way to execute work around legacy infrastructure, including tasks involving customer calls, case files and compliance alerts. The existing core would continue to maintain institutional data and underlying business rules, while an agent could carry out an authorized task and update the bank’s systems.

The approach would start with narrow, repetitive assignments, according to Choudhary. A lending agent could first screen leads, then take on prequalification or full applications if it performs reliably. Banks would set its role and permissions, monitor its performance and determine when work must go to a person or another authorized agent.

For example, Choudhary said a collections agent might process a payment but not reverse one independently. That restriction illustrates the distinction between automating a task and granting an agent authority to make consequential changes. The interview presents this as Kastle’s proposed operating model; it does not report a bank-wide deployment or independently measured results.

AI Could Add Capacity Without Replacing Cores

The proposal matters because replacing a bank’s core system can involve costly, complex changes to infrastructure that supports essential operations. If agents can safely handle selected tasks across those systems, a bank might introduce some new capabilities without waiting for a full core conversion. That is a potential path, not a result established by the interview.

For community banks, the argument is that digital workers could take on routine administrative work while human staff remain available for customer relationships and complicated cases. Choudhary says this could let banks expand lending and service capacity without increasing back-office staffing at the same rate. No quantified staffing savings, productivity gains or customer outcomes were provided in the source material.

The proposal also puts the focus on operational controls, not just model performance. A tool that completes a task in a demonstration may still be unsuitable for a regulated workflow if its actions cannot be checked, traced and managed. Banks would need to judge agents against their own policies, risk limits and service standards.

From Core Replacements to an Action Layer

Banks have long faced the challenge of adding capabilities while keeping complicated, expensive legacy systems running. Choudhary’s argument shifts the starting point: rather than replace those systems first, banks could add agents that interact with them and take on selected work. The core remains the system of record and business logic in this account; the agent performs tasks under defined authority.

That concept treats an AI agent more like a supervised coworker than a standalone software installation. PYMNTS interviewer Karen Webster compared onboarding an agent to onboarding an employee, pointing to roles, access controls, training, reviews and escalation procedures. Choudhary said banks could use similar management structures for digital workers, with permissions expanded only as performance warrants.

The source identifies Kastle as an AI workforce platform for banks and lenders and reports its Series A funding. Choudhary is the company’s co-founder and CEO, so his description is a company executive’s view of the opportunity, not an independent assessment of whether the model is effective across the banking sector.

“We think of the agents as the action layer on top of [legacy cores].”

— Rishi Choudhary, Kastle co-founder and CEO

Deployment Results Remain Unreported

The interview does not identify banks using Kastle agents in production, provide independent performance data or quantify costs compared with core upgrades. It also does not specify how quickly agents could be deployed, how often they would require human intervention, or what safeguards would apply across different banks and tasks.

Choudhary said institutions can configure policies, upload new rules and monitor agent performance through Kastle’s platform. The source does not give details on audit procedures, error rates, security testing, regulatory reviews or responsibility when an agent makes a mistake. It is also unclear whether reducing customer-service calls through better underlying processes is part of the platform’s current capabilities or a broader goal.

Banks Must Test Controls in Practice

The next test is whether banks can connect agents reliably to existing workflows while keeping permissions narrow and actions auditable. A cautious rollout, as described by Choudhary, would begin with routine work and expand only after the agent meets the institution’s performance standards. Banks would also need clear escalation paths for actions outside an agent’s authority.

Further evidence would be needed to compare this approach with core modernization on cost, reliability, customer experience and operational capacity. The PYMNTS interview does not announce a specific bank deployment or timetable, so whether the proposed action layer becomes a practical alternative—or complement—to core upgrades remains open.

Key Questions

Are AI agents meant to replace bank core systems?

Not in Choudhary’s proposal. He describes agents as working on top of legacy cores, which would continue to hold institutional data and underlying business logic.

What work could a bank agent perform?

Examples in the interview include handling customer calls, case files and compliance alerts. Choudhary also described a lending agent that could progress from lead screening to prequalification and application intake as it demonstrates reliability.

Would an AI agent have authority to make any change?

No. Choudhary said banks would define agents’ responsibilities and permissions. His example was a collections agent that could process a payment but could not independently reverse it.

Has the interview shown that AI agents save banks money?

No quantified savings or independently measured results were provided. The interview describes a potential way to add operational capacity, but does not establish its financial impact.

What remains uncertain about the approach?

The source does not provide production deployment details, independent performance data, error rates, implementation costs or a comparison with core upgrades. How agents perform under each bank’s controls remains to be established.

Source: rss