Bayer Crop Science has deployed a new Agentic Decision Engine (ADE) designed to simplify the complex data requirements of its global plant breeding operations. The system serves as a natural language interface over the company’s existing Connected Decision Engine (CDE), allowing researchers, breeders, and data scientists to interact with analytics workflows without manually navigating fragmented operational logs, dashboards, and metadata stores.
Solving operational complexity
Previously, Bayer’s breeding research—which integrates genomics, field trials, and environmental imagery—relied on hundreds of machine learning configurations. As adoption increased, the challenge shifted from executing models to understanding the results. Users faced significant delays when investigating workflow failures or tracing data origins. The ADE addresses this by employing a multi-agent architecture where a supervisor agent routes queries to specialized agents, such as a ‘Troubleshooting Agent’ or ‘Domain Expert Agent,’ which retrieve necessary context and provide answers directly.
According to the company, the engine is resolving approximately 80% of user inquiries without requiring engineer intervention. For instance, diagnostic tasks that once required over an hour of manual log analysis can now be completed in under 10 minutes.
Technical implementation and architecture
The ADE is powered by Amazon Bedrock AgentCore, which provides a framework-agnostic runtime for managing agent sessions and tool access. The architecture includes four distinct layers: a user interface, agent orchestration via LangGraph, specialized domain agents, and a data plane utilizing Amazon DynamoDB and Amazon Aurora. To ensure secure and scalable operations, Bayer integrated AgentCore Gateway to manage authentication and authorization independently from agent code, preventing the need to share tokens across trust boundaries.
The development process involved close collaboration with AWS Unified Operations. Technical Account Managers and generative AI specialists participated directly in the design phase, conducting reviews that targeted specific failure modes and capacity risks. This engagement helped the team transition the system from a prototype to a production-ready environment, incorporating security measures such as role-based access control and structured output validation.
Future directions
Since the rollout, over 100 users across Bayer Crop Science have adopted the platform. The company is currently exploring ways to enable interoperability between agents running in different environments, aiming to allow hybrid cloud communication without manual configuration. By adding an intelligent layer atop its existing analytics infrastructure, Bayer intends to accelerate scientific discovery while maintaining rigorous control over its ML-driven research pipelines.
Source: aws.amazon.com
Original article: https://aws.amazon.com/blogs/machine-learning/how-bayer-crop-science-built-an-agentic-decision-engine-on-amazon-bedrock-agentcore-with-aws-unified-operations/
Leave a Reply