
The agentic AI shortcut to faster, cheaper and safer bank IT overhauls
Piecemeal upgrades are stifling innovation. The path to growth lies in using technology to modernise infrastructure while simultaneously improving the customer experience.
By Mike Gardner, global banking leader, consult partner and vice president at Kyndryl

In mature, global and highly regulated industries such as banking, continuous innovation is the only route to success. However, 58% of financial services leaders say innovation is delayed by foundational tech stack issues.1
Leading banks are facing an unpleasant side effect of the industry’s history of mergers and acquisitions. Both the acquiring and the acquired have often been left with complex legacy architectures that cannot adapt to the demands of modern customers and business requirements. Every year, those fragile environments become more brittle and fewer people have the skills to maintain them.
As banks consider adopting agentic AI technologies, many find the problem is even more acute than suspected: most enterprise infrastructure was never designed for machine-driven operations or tightly governed automation at scale. There are also important security considerations as increasingly powerful AI models become more widely available, with the potential to increase the frequency of attacks while lowering opportunity costs for attackers.
The business implications are too damaging to ignore. An inflexible legacy architecture leads to less agility and greater risk. This means line-of-business leaders become accustomed to hearing “no” when they attempt to innovate, discouraging creative thinking about new ways to bring value to the customer.
Banks that are not able to modernise their tech stacks to use new technologies, such as generative and agentic AI, not only miss out on improved customer experiences and innovations essential for growth. They also miss opportunities to strengthen their cybersecurity.
The risks of inaction—and flawed attempts at modernisation
The debut of Anthropic’s Mythos, an AI model powerful enough to find vulnerabilities in decades-old code,2 marks a turning point in how organisations must approach their IT estate. Organisations accustomed to quarterly cybersecurity updates must develop new approaches to manage a growing volume of patches while maintaining the same rigour within condensed timeframes.
For the finance industry, speed-to-protect will become a critical capability. With the right governance and security frameworks in place, the same agentic AI that can uncover vulnerabilities can also be deployed proactively to defend against AI-driven attacks. This is most achievable in organisations that both recognise the risks of this new era of AI and actively pursue its potential to enhance security and resilience.
For small and medium-sized banks, there are other important considerations: as IT spending continues to rise, it becomes harder for smaller banks to keep up. Without an accelerated transformation, most banks won’t be able to compete with the rapid innovation cycles of modern financial institutions. Mid-sized banks are especially vulnerable, as they can lose customers to credit unions with lower fees and fintechs that can innovate faster because they are not slowed down by old systems.
58% of financial services leaders
say innovation is delayed by foundational tech stack issues
Banks that don’t continuously modernise stand little chance of remaining competitive or independent. In 2025 American banking regulators approved mergers and acquisitions at a pace not seen since 1990.3
However, many banks’ attempts to modernise fall short. Some fail by simply layering AI over old models, others by lifting obsolete processes out of their antiquated tech stack and placing them in a cloud. Neither approach achieves the outcomes most leaders are seeking. These piecemeal, siloed efforts can stretch timelines, heighten risk and compound the bank's technical debt.
The most successful path to achieve value at scale is the application of agentic technology. This can rapidly modernise legacy systems while supporting innovation. A partner that understands an organisation’s enterprise architecture, data and overall infrastructure footprint—including on-premises, cloud and hybrid environments—can deploy the technology to dramatically accelerate modernisation timelines, reduce risk and boost predictability.
Agents built to be “secure by design” can efficiently and cheaply translate legacy code into modern languages, map data dependencies and extract business logic from old code, without disrupting operations. This allows processes to continue to run resiliently while supporting growth. Agile modernisation enabled one European bank to migrate 3,500 workloads with no downtime.
Agentic use cases: building and deploying trustworthy agents
With the appropriate governance model and security tooling in place, Agentic AI can improve a number of business processes while reducing risk, including account opening, onboarding, lending, underwriting, collections and many service functions. Improvements can even be made in areas that are typically the domain of enterprise resource planning (ERP) software, such as auditing or financial operations.

Account opening and onboarding provide persuasive use cases for agents. Every account opening requires a full suite of background checks and research to comply with know-your-customer rules and prevent crimes such as fraud and money laundering. Many banks have already automated some of this repetitive, manual work to speed up onboarding for individual customers.
Onboarding businesses is more complex and prone to error. An estimated 60% of potential customers are lost due to frustrating onboarding processes, research has found.4 Agents can be used to confirm the identities and credentials of anyone trying to open a business account. If the account is a limited liability company (LLC), an agent can find out who owns the LLC and identify the source of their funds. This means onboarding is faster and the risk of customer abandonment plummets.
These processes benefit from a policy-as-code approach—the translation of business rules, compliance requirements and operational policies into guardrails embedded in software.
To learn more about Kyndryl’s approach to agents and modernisation, read A Strategic Guide for Banking Leaders.
In any policy-driven, process-heavy part of the value chain, an agent can ingest the workflow and rules that govern it, ensuring traceability and an audit trail. Policy-as-code helps ensure agents only take permitted actions without hallucinating, giving them the ability to transform regulated workloads. This means any changes to the rules can be made on the fly. Instead of a human coder manually making a change and checking the entire process to make sure it doesn’t break anything, policy-as-code means a single rule update will be applied consistently throughout the process.
Agentic technologies are still in their infancy, but they’ve already shown great promise for the banking industry, especially when deployed with change management to help employees adapt. More than 70% of leaders say their workforce is not able to leverage AI, and 45% say their employees are resistant or openly hostile to it.5 Banks that embrace change management alongside agentic technologies have the best chance of creating a true innovation platform supported by a modern operating model and welcomed by employees.
Using these technologies to accelerate modernisation will be necessary for banks to achieve consistent growth, thrive in an era of AI-driven change, maintain competitive advantage and determine their own futures.
References
- 1.
- 2.
- 3.
- 4.
- 5.