Bayer Crop Science launches agentic decision engine on AWS Bedrock
Bayer Crop Science created the Agentic Decision Engine (ADE), a natural‑language interface that sits on top of its Connected Decision Engine (CDE). ADE uses Amazon Bedrock AgentCore to route user questions through a supervisor agent and up to 22 specialist agents that pull context from tools such as DynamoDB, Aurora PostgreSQL, S3 and OpenSearch. In its first phase the system resolves about 80 %…
Key points
- ADE resolves roughly 80% of user questions without engineer involvement.
- 22 specialist agents handle domain‑specific tasks, cutting troubleshooting time to under 10 minutes.
- Prompt‑prefix caching lowered input‑token cost by 20–30%.
The build relied on three AgentCore capabilities: isolated runtime sessions, IAM‑scoped tool access via AgentCore Gateway, and a framework‑agnostic design that let Bayer bring its own LangGraph orchestration and model choice. Observability was hardened with custom CloudWatch metrics to distinguish task failures from transport success, and cost attribution mechanisms reduced input‑token spend by 20‑30 % through prompt‑prefix caching. Security follows least‑privilege principles, and Bayer plans to evaluate Bedrock Guardrails for additional safety.
AWS Unified Operations supplied a technical account manager and a generative‑AI domain specialist engineer who participated in design reviews, ensuring the architecture met both AWS service behavior and Bayer’s operational needs.
The story so far
2 episodes →- Bayer Crop Science launches agentic decision engine on AWS Bedrockthis story
How Bayer Crop Science built an agentic decision engine on Amazon Bedrock AgentCore with AWS Unified Operations
AWS Machine Learning Blog · 23 October 2026
Loading the full article…
This text was published by AWS Machine Learning Blog and written by Subbu Dandamudi. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Agents & Tools
All →- AI writes 214-page wiki from 45,047 Enron emails, each fact cited · 1 src
- Microsoft adds Work IQ retrieval tool to ground AI agents in enterprise data · 1 src
- Microsoft says its AI diagnostic system outperforms doctors on complex cases · 16 src
- Agentic AI Foundation Announces Four New Projects: Agent Skills, Open Responses, OpenEnv, and OpenSandbox · 7 src
- Microsoft Research releases Flint-Interactive for AI‑generated interactive charts · 1 src
Comments
via GitHub Discussions