The study, titled 'Drop the Hierarchy and Roles: How Self-Organizing LLM Agents Outperform Designed Structures', was published on April 1, 2026, on ArXiv AI.
It involved a computational experiment with 25,000 tasks, utilizing 8 different models and varying the number of agents from 4 to 256.
The research tested 8 distinct coordination protocols, shedding light on the operational efficiency and autonomy of self-organizing multi-agent systems.