Why multi-agent teams undermine expert capability

Recently, Multi-agent LLM systems have been evolving from following predefined workflows toward autonomous collaboration models in which agents freely interact and cooperate with one another. In such environments, efficient collaboration methods cannot be designed in advance; they must emerge naturally through interaction.

Most existing research has enforced collaboration through fixed roles, workflows, or aggregation rules. However, it remains largely unknown just how much genuine synergy Self-organizing teams can achieve when collaboration is unconstrained.

Based on Organizational Psychology theory, this study analyzes whether LLM-based self-organizing teams can achieve real synergy in an unconstrained environment.