Researchers at Weill Cornell Medicine say a new artificial intelligence system designed to work like a collaborative medical team could help speed up one of the slowest parts of drug development: clinical trial design.
The system, called EmulatRx, was evaluated in a study published on July 7, 2026 in Nature Communications. According to the research team, it uses real-world patient data to simulate and improve trial designs, with the goal of making studies faster, more affordable and more precise.
How the system works
EmulatRx is built around five computational agents that mirror a scientific team. A coordinating Supervisor manages the workflow, while separate agents act as a Trialist, Informatician, Clinician and Statistician. Together, they exchange information in natural language, identify problems and revise recommendations during the design process.
The researchers tested the system with de-identified electronic health records drawn from large clinical databases. The data covered both acute conditions, including heart failure, septic shock and kidney injury, and chronic diseases such as Alzheimer’s and Parkinson’s. The records also included diverse populations, including older adults and patients with multiple conditions who are often underrepresented in traditional trials.
Using patient data to replicate trial logic
The team used a method called target trial emulation, which applies the key features of a randomized clinical trial to routine care data. That meant using eligibility criteria, treatment groups, follow-up and outcomes to see whether the system could reproduce known findings and identify differences in treatment effects across patient subgroups.
According to the study, EmulatRx reproduced many previously reported treatment effects across historical trials. The researchers say that could make it useful for testing trial designs before costly studies begin and for spotting when a treatment may help one group while posing risks to another.
Human oversight remains central
The researchers emphasized that the system is not meant to replace scientists or clinicians. Instead, experts can monitor the agents’ exchanges, pause the process, correct decisions and guide the system to reconsider its approach. The team said this human-in-the-loop design is important to keep the system from moving in an unreasonable direction.
Before any clinical or commercial use, EmulatRx will need broader validation across different health systems and patient datasets. Even so, the study suggests that agentic AI could become a useful tool for investigators looking to design better trials with fewer wasted resources.
The work adds to growing interest in AI-supported medical research, especially in areas where trial planning is complex and time-consuming. If further validation confirms the early results, systems like EmulatRx could help researchers move promising treatments toward patients more efficiently while maintaining scientific rigor. More details are available in the original study summary from Weill Cornell Medicine at Weill Cornell Medicine and in the report from News Medical.
