Weill Cornell researchers propose AI agents to flag clinical lab problems earlier

Weill Cornell Medicine researchers have proposed an AI framework to spot trouble across clinical laboratory operations before it delays testing and patient care. Described in a perspective published October 28, 2026, in npj Health Systems, LabARIA would coordinate specialized AI tools rather than replace existing laboratory systems. It remains a conceptual design, not a deployed clinical system.

The proposal addresses a gap the authors see in laboratory AI: individual algorithms tend to handle narrow jobs, with their capabilities separated into task-specific silos. LabARIA—short for Laboratory Agentic Reasoning and Integrative Analytics—is intended to bring those capabilities together. The researchers’ aim is to find patterns across laboratory operations that a tool watching only one part might miss.

A coordinator connects the specialist tools

At the center would sit an AI “data coordinator,” which routes information to agents: specialized AI tools assigned different monitoring jobs. The framework would connect to laboratory information systems and other databases already in use. Those connections are meant to supply the agents with information, without requiring the framework to take the place of the underlying systems.

The proposed division of work spans the condition of incoming samples, the testing process itself and the resources needed to keep a laboratory running:

• Sample quality and speed: agents would monitor specimen quality and testing turnaround times, covering both the material being tested and how long the testing process takes.
• Testing quality and oversight: other monitoring jobs would cover analytical quality control, instrument performance and regulatory compliance, bringing these operational checks into the proposed framework.
• Resources: additional agents would forecast inventory and staffing needs, extending the design beyond the quality of individual specimens or the performance of testing instruments.

Looking for trouble before a rule is broken

Lead author Dr. He Sarina Yang says laboratory staff currently run checks several times a day and investigate potential analytical problems retrospectively when results break established quality-control rules. LabARIA could instead flag concerning trends before a formal violation. One example is a change in testing performance associated with reagents—the materials used in tests—or other testing supplies.

Another proposed use starts before a specimen reaches a testing instrument. Collection, handling or transportation can compromise samples. The researchers say LabARIA could identify patterns in specimen-quality indicators that deserve investigation. Their intended benefit is earlier intervention, helping staff avoid disruptions and delays in reporting results rather than only troubleshooting after a problem has surfaced.

The work targets laboratories that analyze blood, urine, tissue and other bodily samples to support diagnosis and treatment. The researchers describe growing test volumes, workforce shortages and increasingly complex diagnostics as mounting pressures on those operations. Their proposal puts AI around that testing workflow, with laboratory personnel responsible for investigating the problems it identifies.

AI is a good fit for clinical laboratories because of their structured workflow and established quality-control procedures, but human oversight is central to the design

Dr. Fei Wang, senior author and Weill Cornell Medicine associate dean for artificial intelligence

Past laboratory data comes before deployment

The proposed accountability features include traceable data records and explanations of the reasoning behind recommendations. Laboratory professionals would be able to override the system. That gives the human-oversight commitment a specific form: staff would have records and reasoning to examine, as well as the authority to reject a recommendation when necessary.

The next step is experiments using retrospective laboratory data—records of past laboratory activity—to refine LabARIA. Wang says the team would discuss implementing it in its clinical laboratory only after obtaining convincing evidence. The perspective therefore describes an architecture and a testing plan, not results showing that the system has prevented testing delays.

Wang also says monitoring would continue after any deployment to check that the system operates securely and effectively. That ongoing review is part of the researchers’ stated plan, not a description of monitoring already underway in a clinical installation.