Bilevel Coordinated Reflection: A Game-Theoretic Approach to Multi-Agent LLM Systems

Researchers introduced Bilevel Coordinated Reflection, a game-theoretic framework for multi-agent LLM systems that unifies coordination, memory improvement, and external verification. The approach models orchestrator-worker interactions as a Stackelberg game, enabling principled coordination and iterative refinement.

Multi-agent LLM systems commonly use an orchestrator to decompose a task for a team of workers and then improve through textual reflection. Despite strong empirical results, these systems lack a unified account of coordination, memory improvement, and the role of external verification. We model orch