More than 65 equipment-failure tickets arrive every day, and it used to take an average of 17.4 hours to receive a first response. After deploying an AI operations system, Natera cut that time by roughly 130x and automated 96.2% of support tickets. This matters because enterprises are converting the experience of a few senior employees into reusable, scalable operational capabilities.
What This Is
Natera is a genetic-testing company whose laboratories operate a range of precision instruments. Equipment downtime slows sample processing and patient reporting, yet troubleshooting knowledge has long remained in individual senior engineers' hands rather than being standardized into manuals.
Together with AWS, the company built SWARM, an agentic system for laboratory operations. Agents—AI systems capable of autonomously invoking tools and handling tasks step by step—connect to internal systems such as Jira, Confluence, and knowledge bases. They automatically read tickets, retrieve historical experience, diagnose faults, and propose solutions, while access controls, encryption, and operational tracing help meet enterprise security requirements.
SWARM is built on Amazon Bedrock AgentCore. This managed service handles agent deployment, tool connectivity, security, and runtime management, allowing the team to focus on diagnostic accuracy rather than building underlying infrastructure.
Industry View
We believe the Natera case sends a clear signal: AI's value inside enterprises is shifting from “writing documents for people” to “enabling organizational knowledge to participate in work automatically.” More important than faster response times is the fact that tacit expertise—which once risked disappearing when employees left—now remains embedded in workflows and is continuously tested against actual results.
That model still has limits. First, automating 96.2% of cases requires sufficiently high diagnostic reliability. If an incorrect recommendation affects instrument operation, the cost could exceed that of a delay caused by a human. Second, these results came from a medical laboratory: a relatively closed, standardized, high-frequency environment. Enterprises cannot simply copy them. Equipment types, data quality, access-control systems, and accountability structures all differ, making SWARM's numbers difficult to reproduce.
Furthermore, once agents can access tickets, documents, and operational tools, excessive permissions, cascading errors, and responsibility attribution become governance issues. Real-world deployment requires human review, audit trails, and failure-recovery mechanisms—not blanket autonomy simply because the demonstration looks smooth.
Impact on Regular People
For enterprise IT: Enterprise knowledge-base development will shift from “uploading files for AI” to “connecting AI to real workflows and assigning it tasks.” Whether a system can integrate with existing ticketing, equipment, and access-control systems will determine implementation outcomes more strongly than model parameter counts.
For individual careers: Senior employees' expertise will become easier to capture and invoke. This amplifies their capabilities while also pushing organizations to reconsider the value of repetitive operations work. People who can clearly articulate field judgment, exception handling, and accountability will not be easily replaced in the short term.
For consumers: Faster laboratory troubleshooting could accelerate sample flow and report delivery. However, this remains an internal enterprise-efficiency case. Whether consumers will notice a meaningful service improvement depends on how much of the saved time actually reaches the delivery side.