Architectural Design, Not Only Model Intelligence, Governs Multi-Agent LLM Performance

A study showing that multi-agent LLM system performance depends on architectural design choices, not just individual model intelligence.

Auto-scored: This is a neutral research finding about multi-agent LLM design with no direct risk or mitigation implications.