Technology
Why Multi-Agent AI Systems Fail and How Team Structure Matters
A 2025 study led by researchers at UC Berkeley examined over two hundred execution traces across seven open-source frameworks to understand why multi-agent AI systems fail. The analysis revealed that the primary bottleneck is rarely the underlying language model itself, but rather structural issues in specification, agent coordination, and validation.