Researchers have proposed a new method for identifying the causes of failures in multi-agent systems. The existing approaches rely on large language models (LLMs) to perform this task, but these methods are computationally expensive and often inaccurate. In contrast, the new approach uses a lightweight graph-based framework called AFANet, which is able to match or outperform LLM-based baselines with significantly fewer parameters and near-zero inference cost. --- Why it matters: This matters because it suggests that complex tasks like agent failure attribution don't require large language models, but can be achieved more efficiently with simpler approaches. This has implications for the development of multi-agent systems in areas such as robotics and autonomous vehicles. Source: https://arxiv.org/abs/2608.18575