How Bayer Crop Science built an agentic decision engine on Amazon Bedrock AgentCore with AWS Unified Operations

This article describes how Bayer Crop Science built the Agentic Decision Engine (ADE), a multi-agent natural language interface over their Connected Decision Engine (CDE) ML workflow platform using Amazon Bedrock AgentCore.

• ADE uses a supervisor agent that routes questions to 22 specialist agents covering domains like troubleshooting, operations, and visualization
• Architecture separates user experience, agent orchestration, specialist agents, and tools/data sources across DynamoDB, Aurora, ElastiCache, S3, and OpenSearch
• Five key design decisions: externalize orchestration state to DynamoDB, separate identity boundaries for Entra ID and IAM, treat model selection as interface contract, scope tools per agent via AgentCore Gateway, and promote complex capabilities to agents
• Production preparation included task-level observability metrics, cost attribution across models and agents, and security hardening with least-privilege IAM roles
• Since launch, 100+ users resolved 80% of questions without engineer support; prompt caching reduced input-token cost by 20-30%

ADE demonstrates how agentic systems require careful attention to state management, identity boundaries, observability, and cost attribution beyond prompt engineering to achieve production-ready AI applications.