The Organizational Physics of Multi-Agent AI: Substrate-Independent Dysfunction in Autonomous Software Engineering Swarms

This report presents controlled empirical research comparing four multi-agent AI coordination architectures on identical software engineering tasks. We present empirical evidence that organizational dysfunction is substrate-independent. In a controlled comparison, four coordination architectures—single agent, hierarchical, stigmergic (8 concurrent agents), and gated pipeline—built the same 7-service backend using the same LLM and $50 budget. Performance was inversely correlated with coordination complexity: 28/28, 18/28, 9/28, and 0/28. The pipeline consumed its entire budget on planning. The hierarchical coordinator refused to delegate. The stigmergic agents produced incompatible interfaces at every boundary. Only the single agent—with no coordination architecture—succeeded fully. In two additional studies, a pipeline swarm equipped with six explicit anti-dysfunction mechanisms produced the dysfunction those mechanisms were designed to prevent: bikeshedding, governance conflicts, backward pipeline oscillation, and verification theater. A contract-first alternative narrowed the Goodhart gap but introduced specification perfectionism, suggesting dysfunction migrates across architectures but does not disappear. Results are formalized using Crawford–Sobel signal degradation, Goodhart's Law, and the Data Processing Inequality. Coordination failure arises from information-theoretic constraints on any system coordinating through compressed representations—not from properties of the agents.