AWS re:Invent 2025 - Building Predictable AI Agents with Pega and Amazon Bedrock (MAM104)

Summary of "Building Predictable AI Agents with Pega and Amazon Bedrock"

Introduction to Agents and AI

• Presenter is Don Sherman, CTO at Pega, a long-time AWS partner software company
• Pega focuses on automation, workflow automation, and decisioning automation for large enterprises and government agencies
• While enterprises are excited about agents, they are also concerned about the unpredictability of agents operating with 90-95% accuracy
• AI is not a single technology, but a collection of different techniques that solve various problems

Types of AI

1. AI Decisioning: Using structured data to make predictions and decisions, such as determining the best offer for a customer or predicting process SLA misses. Pega uses this in their Customer Decision Hub, and leverages Amazon SageMaker models.
2. Generative AI: Productivity tools like meeting summarization and knowledge retrieval, powered by technologies like AWS Bedrock.
3. Transformative AI: Using AI to rethink and redesign business processes, not just enhance existing ones. Pega's Genai Blueprint is an example of this.

Genai Blueprint: Designing Predictable AI Workflows

• Genai Blueprint uses AI capabilities from AWS, Anthropic, and Pega's own expertise to help enterprises rethink their business processes.
• It creates a "score" or workflow that guides agents to operate in unison, rather than randomly.
• Genai Blueprint can be used to design end-to-end applications, such as an insurance claims management process, by:
  • Identifying the necessary workflows (e.g., first notice of loss, damage assessment, business interruption)
  • Defining the data models and business rules
  • Determining the user roles and security privileges
  • Suggesting the appropriate channels (desktop, mobile, contact center, etc.)

Deploying Predictable Agents

• Genai Blueprint allows users to run a prototype of the designed application, testing the workflows and user experiences.
• Agents can be embedded at different steps of the workflow to automate specific tasks, such as validating photos or signatures.
• This approach allows agents to operate within the defined workflows, rather than acting independently, ensuring predictability and control.

The Future of Agentic Experiences

• The goal is to move away from siloed applications and towards a "fabric" of workflows, knowledge, and data that can be accessed through a single, agentic interface.
• This fabric can be composed of workflows and data from various systems, both within Pega and external to it, enabled by partnerships with AWS, Capgemini, Accenture, and others.
• The result is a more seamless, predictable, and personalized experience for both employees and customers, powered by a network of agents operating within defined workflows.

Key Takeaways

• Pega and AWS are collaborating to enable enterprises to build predictable AI agents that operate within defined workflows and business processes.
• Genai Blueprint uses AI to design and prototype these workflows, ensuring they meet the needs of the business and can be reliably executed by agents.
• The future of enterprise software is moving towards a "fabric" of workflows, knowledge, and data that can be accessed through a single, agentic interface, enabling more seamless and personalized experiences.