Paired experiment on 86 runs: adding a manager tier to an LLM agent team costs 51.5% more tokens, adds 53% more hedging, and lowers utility

arXiv·medium signalXBlueskyLinkedInCopy link

A September 13 arXiv paper ran five-agent business-intelligence reporting with one variable changed, whether a Manager could reject and request revisions, across 43 paired products and 86 runs judged by a five-model panel plus deterministic spec checks. Flat organizations scored higher on Utility (d = 0.42, p = 0.009) and Writing Clarity (d = 0.34, p = 0.030), hierarchical reports hedged 53% more, each revision loop cost 0.14 points of clarity, and the supervisory tier consumed 51.5% more tokens with no quality gain. The authors' rule: a supervisor pays for itself when it can verify and becomes a liability when it can only opine.