How Natera designed an agentic AI system for lab ops with Amazon Bedrock AgentCore

This article describes how Natera built SWARM, an agentic AI system for lab operations using Amazon Bedrock AgentCore to capture and deploy institutional knowledge at scale.

• SWARM reduces first response time from 17.4 hours to 8 minutes (130x faster) and automates 96.2% of support tickets
• Two complementary agents (swarm-agent for autonomous batch processing, swarm-hive for interactive engineering) share a versioned skill library of 33 skills
• Custom document processing pipeline with Amazon Titan Text Embeddings and Amazon Bedrock Knowledge Bases ensures diagnostic accuracy across equipment families
• Keyless deployment using OIDC and AWS STS, IAM least privilege, VPC isolation, and automatic credential rotation secure the system for clinical environments
• HIPAA-compliant architecture with audit logging, no patient data in vector index, and Business Associate Agreement coverage
• Continuous evaluation loop grades diagnoses as Accurate, Partial, or Wrong; high-confidence predictions achieve 74% accuracy
• Per-invocation tracing with AWS X-Ray and Amazon CloudWatch provides observability required for regulated clinical operations

SWARM demonstrates how agentic AI systems can scale institutional knowledge in high-volume operational workflows while maintaining security, compliance, and diagnostic accuracy in healthcare environments.