Toward an Organizational Science of Multi-Agent LLM Systems: Decoupling Who, How, and Which Algorithm

The paper introduces IMACS, a multi-agent LLM framework that separates three commonly entangled dimensions: team organization, coordination, and the algorithmic protocol used to combine agent work. Organizational theories including Belbin roles, Mintzberg coordination, and RACI accountability become executable configuration factors. Six published collaboration algorithms share a common interface, enabling controlled comparisons in which organization changes while the protocol remains fixed. An Adaptive Org Routing contextual-bandit meta-protocol selects protocols under an explicit quality-cost objective and learns online from benchmark and LLM-judge rewards. Ablations report that accountability matters when routing passes through the accountable agent, while the best placement changes across model families.