This partial HN capture records a debate about multi-agent coordination: commenters describe quoted failure modes, argue over hierarchy and information sharing, and offer contrasting accounts of their own agent setups.
Why it matters: The comments suggest treating multi-agent design as a tradeoff to test: splitting information or using unconstrained peer agents may not help, while explicit roles, specialist tools, and review rules may improve a particular workflow. These are commenters’ arguments and experience reports, not general results.
StateM uses durable state, checked transitions and runbooks; the authors report Terminal-Bench results across several models and a separate BusinessBench evaluation.
Why it matters: Execution controls can improve task completion without changing the base model.