Clinical trial design is time consuming and requires substantial domain expertise. Real-world data such as electronic health records encodes practice-based evidence of great value for trial design. Machine learning methods can extract this evidence, but they still require extensive iteration with domain experts before becoming useful. We introduce EmulatRx, an agentic AI framework that derives real-world evidence for trial design. Through iterative conversation and analysis across agents with different roles, EmulatRx autonomously refines trial protocols and generates a report with actionable insights. We applied it to acute diseases using MIMIC-IV and chronic diseases using the INSIGHT Network across five New York City health systems, demonstrating its ability to facilitate and accelerate trial design.