By Nabeel A. Siddiqui, MSMS (MIT Sloan), MBA (HEC Paris), BEngg (Computer Science)

https://www.linkedin.com/in/nabeelsiddiqui

In the high-stakes world of global finance, the greatest threat to innovation isn’t a lack of vision; it is the weight of the past. Major financial institutions are currently powered by millions of lines of COBOL code, much of it housed on mainframes that have been operational for over forty years. Until recently, the only solution to this “technical debt” was a “rip-and-replace” strategy, an endeavor that typically costs billions, takes a decade, and carries an immense risk of systemic failure.

However, 2025 has introduced a third path: Agentic AI.¹ Rather than attempting to dismantle legacy infrastructure, banks are deploying autonomous agents as a “cognitive glue.”² These agents act as an intelligent layer that sits atop legacy cores, translating archaic data into modern insights, automating complex cross-system workflows, and even refactoring old code in real-time.³

1. Beyond the Chatbot: Why “Agentic” Matters

While the first wave of AI in banking focused on generative assistants (chatbots that answer questions), Agentic AI focuses on execution.⁴ An agent is goal-oriented; it has the reasoning capability to break a complex objective—such as “onboard a corporate client”—into a series of autonomous steps.⁵

In a legacy banking environment, data is often trapped in silos.⁶ A simple request might require pulling data from a 1980s mainframe, cross-referencing it with a 2010s CRM, and validating it against a modern cloud-based compliance database. In the past, this “swivel-chair” task required a human. Today, agentic systems use multi-agent orchestration to navigate these disparate systems independently, acting as the connective tissue that makes a 40-year-old bank feel like a modern fintech.⁷

2. Publicly Acknowledged Successes

BNY: The “Eliza” Ecosystem

BNY, a 241-year-old institution, has set the global benchmark for agentic transformation.⁸ The bank built its own proprietary enterprise AI platform, Eliza, which now orchestrates over 125 AI-enabled solutions.⁹

What makes BNY’s approach truly “agentic” is the deployment of “Digital Employees.” These are not just algorithms; they are software agents with their own bank user IDs, email accounts, and system permissions.¹⁰ These agents work side-by-side with human payment teams to monitor market conditions and detect risk signals in real-time.¹¹ By harnessing advanced reasoning, BNY’s agents can benchmark negotiated contracts against corporate best practices and regulatory requirements, a task that once required hours of manual legal review.¹² As of late 2025, 99% of BNY staffers have been trained on these tools, and over 20,000 employees are actively building their own agents to automate niche legacy workflows.

HSBC: NOLA 2.0 and Risk Advisory

HSBC faced a classic legacy challenge: its on-premises hardware for traded risk management was at capacity.¹³ Running “what-if” scenarios for global market exposure took risk managers nearly four hours per simulation.

To solve this, HSBC developed NOLA 2.0 in collaboration with Google Cloud.¹⁴ This agentic framework uses autonomous data-processing agents to spin up cloud infrastructure on the fly, execute trillions of calculations, and shut the infrastructure back down. The results were transformative: calculation speed increased 10x, and scenarios that previously took hours now yield results in just 15 minutes. These agents allow HSBC to perform intra-day portfolio re-pricing and active hedging, giving them a competitive edge in volatile markets.¹⁵

Crédit Agricole: Reducing Back-Office Friction

In Poland, Crédit Agricole Bank Polska turned to Agentic AI to solve the problem of document-heavy legacy processing. By deploying agents to handle the extraction and verification of data from various loan application documents, the bank cut processing times by 50%. This intervention saved more than 750 hours per month, allowing human employees to shift their focus from data entry to high-value customer advisory roles.

3. The Efficiency Dividend: Modernizing without Moving

The business case for Agentic AI as “glue” is supported by staggering productivity metrics.¹⁶ According to research from McKinsey & Company (2025), agentic deployments in core operations can unlock up to 40% productivity gains.

Furthermore, Deloitte predicts that Agentic AI will help the banking industry save between 20% and 40% in software investments by 2028, as firms choose to “wrap” their legacy systems in AI rather than replacing them.

4. The “Human-on-the-Loop” Security Model

Modernizing legacy systems via AI introduces new risks, particularly around data privacy and “agent drift.”¹⁷ To combat this, banks are adopting a “Human-on-the-Loop” architecture.

For example, at BNY, while agents can autonomously detect and fix issues in legacy code, a human engineer must still approve the final change before it is deployed to production.¹⁸ This ensures that while the “heavy lifting” is autonomous, the final accountability remains human.¹⁹ Additionally, banks are deploying “Compliance Agents” whose sole purpose is to monitor other agents, ensuring they follow regulatory guardrails and maintain a clear audit trail.

5. Conclusion: A New Era of “Elastic” Banking

Agentic AI has effectively ended the era of the “all-or-nothing” legacy migration. By acting as a sophisticated, reasoning interface between the old world and the new, these agents allow banks to innovate at the speed of a startup while maintaining the stability and security of a century-old institution.

For the CIO of a global bank, the priority is no longer just “migrating to the cloud.” It is about building a robust ecosystem of agents that can breathe new life into existing assets.²⁰ As the successes of BNY and HSBC demonstrate, the future of banking isn’t found by deleting the past—it’s found by giving the past a brain.

References

1. BNY (2025). Unlocking Value with BNY’s Enterprise AI Platform: The Eliza Case Study.
2. McKinsey & Company (2025). The State of AI in 2025: Agents, Innovation, and Transformation in Banking.
3. Google Cloud (2025). HSBC Case Study: Reimagining Traded Risk with NOLA 2.0.
4. Dell Technologies (2025). BNY: How a Legacy of Innovation Drives Modern Transformation.
5. Kore.ai (2025). Agentic AI in Banking: Transforming Intelligent Automation and Customer Experience.
6. Deloitte Insights (2025). 2026 Banking and Capital Markets Outlook: The Agentic Shift.
7. The Financial Brand (2025). How BNY is Banking on Innovation in AI and Digital Assets.