CollabSim: Measuring Why Multi-Agent LLM Systems Fail at Collaboration

arXiv·medium signal

CollabSim is a CSCW-grounded methodology for controlled multi-agent experiments, built on the finding that LLM multi-agent systems often falter not from weak individual task ability but from poor collaborative competence — establishing common ground, maintaining shared task understanding, and repairing misalignment. It provides a way to isolate and measure coordination failures through text channels. Useful for anyone debugging why their agent teams underperform.