How Bayer Crop Science built an agentic decision engine on Amazon Bedrock AgentCore with AWS Unified Operations
aws ·
Bayer Crop Science developed the Agentic Decision Engine (ADE) to simplify access to its Connected Decision Engine (CDE) by providing a natural language interface for plant breeding data. Built on Amazon Bedrock AgentCore, ADE utilizes a multi-agent architecture where a supervisor agent routes queries to specialized agents that retrieve context from various operational systems. This approach allows users to diagnose workflow failures and understand model outputs without manually navigating complex logs or dashboards. The system successfully resolves approximately 80% of user questions without requiring engineer intervention, significantly reducing troubleshooting time. Key design decisions included externalizing orchestration state to DynamoDB, separating identity boundaries, and scoping tool catalogs to optimize token usage. AWS Unified Operations supported the team with technical guidance, ensuring robust observability, cost attribution, and security readiness before launch. Currently serving over 100 users, ADE enhances R&D efficiency while maintaining strict governance over the underlying analytics platform.
aws