AWS re:Invent 2025 - Building Prudential’s microagent platform with MCP and A2A on AWS (IND3302)

Building Prudential's Microagent Platform with MCP and A2A on AWS

Key Challenges Faced by Financial Services Enterprises

• Enterprises in the financial services industry have ambitious goals to leverage AI and intelligent technologies across various use cases:
  • Intelligent document processing
  • Data analysis using agents
  • Workflow simplification for customers
• However, enterprises face several challenges in scaling these AI initiatives:
  • Developers and teams want flexibility and autonomy to rapidly deploy AI solutions
  • Need to align these solutions with enterprise-wide business goals, constraints, and brand governance
  • Entanglement of agent implementations and shared components leads to blurred ownership boundaries
  • Difficulty in properly managing sensitive data (PII, PHI) across multiple agent solutions
  • Ensuring solutions can adapt to the latest AI/ML trends and maintain optimal performance

Modular Architecture as a Solution

• To address these challenges, a modular architecture with clear separation of concerns is crucial:
  • Separate components for business logic, runtime execution, governance, and scaling
  • Identify key components like runtime, memory, code interpreter, and browser tools
  • This allows for easy swapping of components and adaptability to new trends

Credential's AI-Powered Advisor Assistant

Business Challenges and AI Opportunities

• Credential Financial, a large financial services company, faces challenges across its business units:
  • Fragmented advisor workflows and limited personalization in distribution and sales
  • Manual, slow underwriting and rigid risk models in underwriting and risk
  • High servicing costs and lengthy claim processes in customer service and claims
  • Slow innovation cycles in product development and generic campaigns with low conversion rates in marketing
• AI can help address these challenges by enabling:
  • Hyper-personalized advice and sales enablement
  • Real-time risk assessment
  • Self-serviceable processes and automated claims triage
  • AI-driven market insights and personalized product offerings

Life Insurance Advisor AI Assistant

• Advisors typically perform a series of manual, time-consuming tasks across the client engagement lifecycle:
  • Client engagement, needs assessment, solution design, product presentation and illustration, application and underwriting support, and ongoing policy servicing
  • Advisors have to navigate multiple IT systems and lack a unified, context-aware experience
• Credential's AI-powered Advisor Assistant aims to address these challenges:
  • Provides a conversational, natural language-driven interface for advisors to access various functionalities
  • Orchestration agent routes requests to specialized sub-agents for quick quotes, forms, product information, illustrations, and book of business management
  • Quick Quote agent uses trained models to provide instant underwriting decisions based on advisor's questions about client medical conditions

Scalable and Secure Microservices Architecture

Key Design Principles

• Enable scaling with time by:
  • Adapting to the latest AI/ML trends, frameworks, and technologies
  • Maintaining optimal performance through model, prompt, and implementation updates
• Democratize solution development and deployment across the enterprise:
  • Standardize frameworks, pipelines, tooling, and tech stack
  • Minimize platform team involvement for new agent deployments

Architecture Overview

• Secure user authentication and context management:
  • SSO-based authentication with user access control
  • Secure token and context-specific window ID for session management
• Modular agent architecture:
  • Orchestration agent coordinates requests and responses between UI and specialized sub-agents
  • Sub-agents (e.g., Quick Quote, Forms, Product) leverage LLM gateway and knowledge management for responses
  • Agents use MCP, A2A, and React-based frameworks for execution and context sharing

Platform-based Approach

• Developed a centralized Agentic AI Platform to address scalability and reusability challenges:
  • Provides common services like vector store, agent runtime, LLM gateway, and enterprise DevSecOps tooling
  • Enables responsible AI practices, faster time-to-value, and improved monitoring/observability
• Key benefits of the platform-based approach:
  • Reduced turnaround time for new use case deployment (from 6-8 weeks to 3-4 weeks)
  • Ability to quickly integrate new agent capabilities based on user feedback
  • Reduced technical debt by centrally managing platform upgrades and changes

Lessons Learned and Future Directions

Lessons Learned

• Agents are not suitable for all business problems; some use cases require standalone LLM applications
• Solutions need to be built with an end-to-end value chain perspective, not just individual components
• Unpredictable drops in agent performance due to model upgrades and edge cases in training data
• Importance of standardized approaches for memory management, observability, and context engineering

Future Directions

• Integrating with agents from other line-of-business systems to enable cross-functional reuse
• Developing a standardized "Agent Development Layer" on top of a "Core Platform Layer":
  • Agent Development Layer: Enables rapid agent building and deployment by data scientists, engineers, and AI enthusiasts
  • Core Platform Layer: Provides foundational services like interpreter, execution, context engineering, and agent management
• Leveraging a centralized registry and performance monitoring for better agent discoverability and management

Key Takeaways

1. Scalability with time is crucial and requires a modular architecture that separates concerns and enables easy component swapping.
2. Credential's AI-powered Advisor Assistant demonstrates how AI can transform financial services workflows, enabling personalized advice, streamlined processes, and improved decision-making.
3. A platform-based approach with standardized frameworks, tooling, and centralized services can significantly improve time-to-value, reusability, and manageability of AI solutions at scale.
4. Integrating agents across different business functions and maintaining robust context engineering are key challenges that require careful architectural design.
5. Enterprises should adopt a layered platform strategy, with a core foundation for common services and a flexible agent development layer to democratize AI solution building.