This session explores how S&P Global redefined AWS rightsizing by embedding Agentic AI directly into its FinOps operating model. Rather than relying on static recommendations, we built an autonomous, policy-aligned rightsizing workflow where AI agents continuously analyze real-time telemetry, FinOps data, and observability signals to make context-aware decisions—while keeping humans in the loop for governance and safety.
Attendees will learn how this approach transforms AWS rightsizing from a manual, time-intensive process into a continuous, scalable, and governed system. We’ll walk through the architecture behind the solution, including multi-agent orchestration using Amazon Bedrock, custom Model Context Protocol (MCP) tools, and human-in-the-loop approval workflows. The discussion will also cover how this agentic framework is being extended beyond EC2 to services such as Amazon RDS, AWS Lambda, and Amazon ECS on Fargate—each with distinct scaling behaviors and cost models. Finally, we’ll share lessons learned around trust, governance, and observability, including how LLM observability enables transparency, auditability, and continuous improvement of AI-driven decisions.