Cognitive Heterogeneity and Behavioral Biases in Multi-Stage Supply Chains: Evidence from LLM-Based Agent Simulations

Modeling cooperation and coordination among LLM-based agents in multi-round decision making environments is a central challenge in artificial intelligence and operations management. Traditional behavioral experiments have demonstrated that cognitive biases can generate systemic inefficiencies, but such approaches are costly, difficult to scale, and limited in their ability to control individual heterogeneity. We propose a new experimental paradigm that employs LLM-based agents to simulate multi-stage supply chain decision making. The results indicate that LLM agents exhibit myopic and self-interested behavior that amplifies supply chain inefficiencies, while information sharing emerges as an effective mitigating mechanism. This study extends behavioral operations research by demonstrating how LLM-based agents can both reproduce and illuminate emergent coordination failures, highlighting their potential and limitations as tools for studying complex human like decision making in supply chains.