"The Organizational Physics of Multi-Agent AI" documents a controlled experiment featuring multi-agent AI systems
United States, 26th Feb 2026 —A new study, The Organizational Physics of Multi-Agent AI, reveals that agentic artificial intelligence (Agentic AI) “swarms” recreate the dysfunction in middle management that the technology purports to reduce. The result of a controlled experiment that compared single-agent, hierarchical multi-agent, stigmergic multi-agent, and pipeline architectures on identical software engineering tasks, the report demonstrates how Agentic AI replicates many of the counterproductive processes that are inherent to organizational management.
Jeremy McEntire, a veteran software engineer and engineering executive, undertook the study as part of his analysis of agentic AI as a corporate technology. “The findings here are not encouraging,” explained McEntire. “However, they are not surprising. AI is essentially backward-looking. Its intelligence is based on data from existing management procedures, so it is not a big shock to see that the agents start to act similarly to flawed human beings.”
Specifically, the study showed the following results:
- Single agent: 28/28 tasks completed
- Hierarchical agents: 18/28
- Stigmergic agents: 9/28 tasks completed
- Pipeline: 0/28 tasks completed. The agents consumed the entire $50 compute budget on planning, but produced no deployable code
Review stages rejected 87% of agent submissions. Four rejections had zero factual basis. The pipeline architecture reproduced bureaucratic dysfunction with no humans involved. The research formalizes this usingCrawford-Sobel signal degradation theory, Goodhart's Law, and the Data Processing Inequality.
The findings underscore the importance of diligence for organizations that want to deploy agentic AI. The technology is now the subject of extensive hype, and many enterprises are embracing it without fully understanding its limitations and potential to perpetuate negative management behaviors. As McEntire put it, “If you focus on organizational theory, you should be circumspect, if not outright skeptical about claims you’re hearing about agentic AI. In my experience, organizations fail structurally, not personally. Agentic AI picks up on this through its data ingestion and learning cycles—it mimics the incompleteness, dysmemic pressure, and management layers.”
For more information, visit https://cageandmirror.com
About Jeremy McEntire
Jeremy McEntire is an engineering executive who has served in management roles at Twilio, Wander, and other technology companies. He is the author of seven books on technology and management, including Privacy: The Architecture of Forgetting, a book on cryptographic architecture for privacy-preserving internet infrastructure. His other books include Beyond Code — software engineering judgment and organizational context, Uncommon Leadership— leadership frameworks, and Organizational Physics — systems thinking applied to organizational design.
Media Contact
Organization: Jeremy McEntire
Contact Person: Jeremy McEntire
Website: https://cageandmirror.com
Email: Send Email
Contact Number: +14154041188
Country:United States
Release id:41907
The post New Study Reveals that Agentic AI Swarms Recreate Middle Management Dysfunction appeared first on King Newswire. This content is provided by a third-party source.. King Newswire is a press release distribution agency. We do not accept any responsibility or liability for the accuracy, content, images, videos, licences, completeness, legality, or reliability of the information contained in this article. If you have any complaints or copyright concerns related to this article, please contact the company listed in the ‘Media Contact’ section above.
