Agentic AI: Revolutionizing FSI Regulatory Technology Software Development Lifecycle

The banking industry is on the cusp of a transformative shift, driven by the rise of Agentic AI. No longer a futuristic concept, Agentic AI—enabling AI systems to make autonomous decisions and execute tasks—are poised to redefine how financial services operate. This evolution goes beyond traditional AI and even generative AI, representing a paradigm shift with profound implications for efficiency, customer experience, and competitive dynamics.

We are already witnessing trends pointing toward this transformation. A recent report from Citi highlighted a 17-fold increase in mentions of Agentic AI in corporate documents and press articles in 2024, with expectations of explosive growth in 2025. The financial services sector is leading the charge in AI adoption. Early implementations are emerging across the industry:

• Compliance and Regulatory Functions: Agentic AI is streamlining complex processes like sanctions screening and Know Your Customer (KYC) verification, significantly reducing manual effort and improving accuracy.
• Wealth Management: Personalized, 24/7 guidance is becoming a reality through AI-powered agents, augmenting the capabilities of human advisors and catering to evolving customer needs.
• Corporate Treasury: Agentic workflow is automating cash forecasting and optimizing working capital management, freeing up human resources for strategic decision-making.

These examples, while promising, are just the tip of the iceberg. The true power of Agentic AI lies in the ability to handle complex, multi-step processes autonomously. A compelling example of this potential can be seen in the innovative work undertaken by GXS Bank in Singapore—a leading Southeast Asian superapp—leveraging Snowflake’s Data Cloud.

Overhauling RegTech With Agentic AI

The financial services industry (FSI) is under constant pressure to adapt to a rapidly evolving regulatory landscape. Traditional RegTech software development lifecycles (SDLCs) often prove slow, cumbersome, and expensive, struggling to keep pace with new regulations. Agentic AI is poised to revolutionize how FSI institutions develop and deploy RegTech solutions.

Traditional RegTech development is frequently plagued by inefficiencies. Interpreting complex regulations, generating relevant test data, developing solutions, and conducting thorough testing are sequential, time-consuming processes. This can lead to significant delays, increasing the risk of non-compliance and hindering innovation. Agentic AI offers a compelling alternative by automating and streamlining these critical steps.

GXS: Leading the Way in Agentic AI RegTech

GXS Bank is pioneering the use of Agentic AI on Snowflake to revolutionize its RegTech solution development lifecycle. Imagine a scenario where regulatory changes necessitate a new compliance solution. Traditionally, this process could take months, involving extensive development, testing, and deployment. GXS is transforming this by enabling a team of specialized agents built on Snowflake. These agents can autonomously:

1. Interpret Regulations – One agent specializes in understanding and interpreting complex regulatory documents.
2. Generate Test Data – Another agent creates realistic and comprehensive test data based on the specific regulatory requirements.
3. Develop Solutions – A third agent uses the interpreted regulations and test data to generate a fully functional RegTech solution.
4. Automate Testing – Finally, another agent rigorously tests the generated solution against defined criteria.

Rather than exporting terabytes of data into Amazon S3 and managing transient buckets, GXS leverages Snowflake’s secure data sharing and external function framework to stream training data directly into SageMaker training jobs. Credentials are managed via AWS Secrets Manager, and SageMaker notebooks or pipelines pull data straight from Snowflake into the ephemeral storage of the training instance—eliminating intermediate steps and reducing operational complexity.

Once models are trained, GXS orchestrates deployments via Amazon SageMaker Pipelines. Each Agentic AI component—whether it’s a mapping agent interpreting new regulations or a QA agent validating outputs—can be packaged as a SageMaker endpoint or batch transform job. Autoscaling policies ensure that when demand spikes, additional inference capacity spins up automatically; when demand wanes, resources scale back, keeping costs in check.

The result? A fully fledged, tested RegTech solution in a matter of days—drastically reducing the SDLC from multiple quarters to just days. This speed and agility provide GXS with a significant competitive advantage, enabling them to respond quickly to regulatory changes and offer innovative solutions to their customers.

“The opportunity for agentic AI to tackle the complexity of financial services is like nothing I’ve seen in all my years in the industry. Innovative companies like GXS, the digital bank of Grab, will leap ahead, powered by Genesis Enterprise AI data agents running natively and securely in Snowflake.” Matt Glickman, Co-Founder & CEO, Genesis Computing

Internal Document

Agentic AI in RegTech: A Deep Dive

What makes Agentic AI unique? Their ability to operate in a continuous loop, analogous to how humans learn, adapt and acquire new skills.

First, they think, processing all the information and context they’ve been given. Then, they plan, deciding the best way to tackle the task at hand. Next, they act, using whatever tools, APIs, or interfaces are available to execute their plan. Finally, they reflect, evaluating the results and adjusting their approach for next time. This feedback loop is what makes them so powerful – they’re constantly learning and improving.

Just like a team of humans, there are different types of Agentic AI with different strengths:

• Simple Reflex Agents: These are like rule-based systems, reacting to triggers.
• Model-Based Agents: They use memory to guide their decisions, giving them a bit more context.
• Goal-Based Agents: These agents focus on achieving specific outcomes.
• Utility-Based Agents: They weigh options and make trade-offs, trying to maximize efficiency.
• Learning Agents: These are the ones that continuously improve their performance over time.

You wouldn’t run a business with just one type of human employee, and the same goes for Agentic AI. You need a mix to get the best results. and how you structure these agents is just as important as what they can do. Various structures include::

• Single Agents: These are like task-specific assistants, good for focused jobs.
• Multi-Agent Systems: This is where agents coordinate and collaborate, working together on complex projects.
• Human-Machine Teams: This is where agents work closely with humans, enhancing our abilities and providing support.

Graphics: ByteByteGo

So, how does this work in practice? This translated to specialized agents working in concert to automate key stages of the SDLC. Grab’s approach illustrates the power of a multi-agent system, where different bots collaborate to achieve a common goal. GXS’ RegTech solution on Snowflake leverages a team of specialized agents:

• The Mapping Agent: This agent acts as the bridge between raw regulatory data and actionable specifications. It accepts a government exhibit or regulatory document as input and generates a mapping document—translating regulatory language into a structured format suitable for downstream processing. Crucially, this agent is designed for both automated and human-assisted workflows, enabling human oversight in highly regulated environments.
• The Code Generation Agent: This agent takes the mapping document produced by the first agent and generates DBT (Data Build Tool) code, automating data transformation for regulatory reporting. Manual coding efforts are reduced significantly, ensuring consistency and accuracy.
• The QA and Remediation Agent: Responsible for quality assurance and validation, this agent tests the generated DBT code using synthetic or provided data (or both), identifying and remediating common issues. Like the mapping agent, it allows for human override, providing a safety net for complex or edge-case scenarios.

Ultimately, Agentic AI creates intelligent partners that can help us tackle complex problems, automate tedious tasks, and unlock new possibilities. It’s about building a future where humans and machines work together seamlessly, each leveraging their unique strengths.

Snowflake: Enabling the Agentic AI Revolution

Snowflake’s Data Cloud is playing a crucial role in enabling banks like GXS to realize the potential of Agentic AI. By providing a unified platform for data storage, processing, and governance, Snowflake simplifies the development and deployment of data and AI solutions. This translates to:

• Expedited Time to Value: Banks can rapidly develop and deploy Agentic AI solutions, accelerating time to market and gaining a competitive edge. Snowflake achieves this through its embedded AI stack called Cortex, which abstracts implementation complexity and democratizes AI for users and engineers alike.
• Scalability and Flexibility: Snowflake’s cloud-native architecture offers the scalability and flexibility needed to handle the demands of Agentic AI applications.
• Enhanced Data Governance: Snowflake’s robust governance capabilities ensure that Agentic AI operates on trusted, reliable data, minimizing risks and ensuring compliance.

The Future of RegTech With Agentic AI

While challenges like governance and cybersecurity remain, the outlook for Agentic AI  in banking is bright. We can expect increased investment in infrastructure to support these advanced systems, as the technology’s ability to autonomously handle complex, multi-step processes revolutionizes both back-office operations and customer-facing services. The transformation will not only drive efficiency gains but also foster new business models and opportunities.

Ultimately, a collaborative approach—where human professionals work alongside agents—will be key to unlocking Agentic AI’s full potential and shaping the future of financial services.