Why do multi-agent systems fail despite individual capability?
Multi-agent systems show lower performance than individual models despite coordinating multiple reasoning instances. What structural failures emerge when multiple LLMs deliberate together, and what ecosystem conditions are required for effective autonomous cooperation?
Multi-agent performance is primarily determined by token spending. AI agents are passive by design, not by capability Current AI agents complete only 30 percent of workplace tasks autonomously. Autonomous LLM agents exhibit four distinct failure modes Multi-agent systems fail across 14 distinct empirically grounded failure modes.