We have all sat through the demos. Over the last two years, the narrative around AI in banking has been dominated by “copilots”-assistants that can summarize a 50-page regulatory PDF or draft a polite email to a client. While useful, for those of us deep in the operational trenches of the First Line of Defense (1LoD), there has always been a lingering question – Is that it?
Summarizing text is helpful, but it doesn’t solve the crushing volume of false positives in transaction monitoring. It doesn’t autonomously verify a corporate structure across three jurisdictions during onboarding.
This is where the conversation is finally shifting-from Generative AI, which creates content, to Agentic AI, which executes actions.
We are seeing this shift accelerate rapidly, not just in New York or London, but right here in the Global Capability Centers (GCCs) and domestic banks of India. The First Line of Defense is evolving from a reactive gatekeeper into a proactive, digital workforce.
The Shift from “Chatting” to “Doing”
To understand the magnitude of this shift, we need to look at the unit economics of risk. Banks typically allocate 10-15% of their headcount solely to KYC/AML activities, yet the industry still misses a vast majority of financial crime flows. The problem isn’t a lack of human effort; it’s a lack of scalable autonomy.
Agentic AI changes the equation because it doesn’t wait for a prompt. It possesses “agency.”
Imagine a scenario in commercial lending or onboarding. In the generative era, an analyst might ask a Generative AI model to draft a credit memo. In the agentic era, we deploy a “squad” of specialized agents-a concept emerging leaders like DBS Bank India are already putting into practice to collapse onboarding timelines from days to minutes.
These squads operate like a digital factory where –
- A Data Agent extracts metadata from ID documents using computer vision.
- Research Agent hits global registries to verify Ultimate Beneficial Owners (UBOs).
- Risk Agent scores the entity against dynamic risk appetites.
- A Critic Agent reviews the work for logic gaps before a human ever sees the file.
JPMorgan Chase is already moving toward this “agentic” model for complex, multistep tasks, aiming to equip its workforce with assistants that do more than just chat-they act.
Global Adoption, Local Relevance
What is particularly exciting for this community is India’s role in this transformation. We aren’t just the back office processing the alerts anymore; we are becoming the architects of the agents that resolve them.
We are seeing Indian financial institutions and GCCs leverage this technology to tackle the specific messiness of our market-diverse languages, fragmented data, and high volumes. For instance, DBS Bank India utilized AI-driven document processing to cut data entry workloads by nearly 50%, freeing up human talent for relationship building.
This moves us away from the “periodic review” model-where we annoy customers every three years for updated documents-to Perpetual KYC (pKYC). Agentic systems monitor data streams in real-time. If a director changes or a sanctions hit occurs, the agent wakes up, verifies the change, and updates the profile instantly.
The Trust Equation – Building the “Glass Box”
Of course, the immediate pushback from any CRO reading this is “Governance.” If an agent declines a transaction, “the black box told me to” is not a defensible argument to the RBI or the Fed.
This is where the technology has matured significantly. We are moving toward Agent Decision Records (ADR)-immutable logs that capture not just the outcome, but the reasoning chain of the agent. It turns the “Black Box” into a “Glass Box.”
We are seeing architectures where agents communicate via “intent streams”-publishing their goals and actions to an event broker (like Kafka) so that every step is audible, traceable, and reversible. This ensures that while the agents are autonomous, they remain strictly governed.
Final Thoughts
The future of the 1LoD isn’t about replacing bankers; it’s about establishing a “dual workforce.” The agents handle the high-volume, high-speed data processing-slashing false positives by up to 80%-while our human experts focus on complex investigations and strategic risk decisions.
If you are interested in the architectural blueprint for this transformation-how to actually build the squads, the governance layers, and the data fabric- you can explore this white paper – Reimagining Banking Functions – How Generative and Agentic AI are Shaping the Future.
