A new arXiv paper reports a paired experiment comparing flat and hierarchical coordination in multi-agent LLM teams, where a Manager agent can review and return worker output for revision. The authors draw on classical organizational theory to examine how such loop-back authority affects team performance. The work is announced as a cross-listing on arXiv cs.AI.
A newly announced arXiv paper argues that compound AI systems typically coordinate their multiple LLM workers by adding a higher-level meta-agent, which centralizes task routing, final answers, and stopping decisions in a single expensive component. UnitBoost instead replaces that supervising model with a merge operator for combining worker outputs. The work is listed under cs.AI with a cross-listing in cs.CL.
A newly surfaced research write-up introduces the idea of procedural graphs, in which an AI agent's execution flow is represented as a structure that can modify itself while tasks are being carried out. The approach is positioned as an alternative to fixed, hand-designed pipelines for orchestrating LLM agents. Details come from a project page shared on Hacker News and have not been independently verified.