This research introduces a novel approach to medical AI agents that dynamically adapt their diagnostic reasoning process instead of following rigid workflows.

• Develops flexible medical agents that search for effective reasoning paths rather than using static, predefined steps
• Demonstrates superior performance in skin disease diagnosis compared to traditional fixed-workflow agents
• Leverages reinforcement learning to help agents discover optimal diagnostic strategies
• Creates agents that can better handle diverse clinical scenarios and emerging medical conditions

This advancement matters because it represents a shift toward AI that can truly mimic physicians' adaptive reasoning rather than following simplistic flowcharts, potentially improving diagnostic accuracy and clinical decision support in real-world medical settings.

Learning to Be A Doctor: Searching for Effective Medical Agent Architectures