Do More Agents Help? Controlled and Protocol-Aligned Evaluation of LLM Agent Workflows 事件

Do More Agents Help? Controlled and Protocol-Aligned Evaluation of LLM Agent Workflows arXiv:2606.05670v1 Announce Type: new Abstract: Does adding more agents help an LLM workflow once compared systems share the same benchmark loader, tool access, answer contract, usage accounting, and trajectory logging? We introduce BenchAgent, an evaluation framework that places single-agent, fixed multi-agent (MAS), and evolving MAS workflows under one normalized execution and logging protocol. BenchAgent e