Turns out, putting a cheap AI under an AI 'boss' actually makes the work worse unless the boss is way, way smarter than the worker.
March 30, 2026
Original Paper
Can AI Models Direct Each Other? Organizational Structure as a Probe into Training Limitations
arXiv · 2603.26458
The Takeaway
While we often assume more organization helps, this study found that an AI management layer often acts as 'pure overhead' that fights against the way AI is trained. It turns out that for AI, much like humans, middle management without a clear capability gap just gets in the way.
From the abstract
Can an expensive AI model effectively direct a cheap one to solve software engineering tasks? We study this question by introducing ManagerWorker, a two-agent pipeline where an expensive "manager" model (text-only, no code execution) analyzes issues, dispatches exploration tasks, and reviews implementations, while a cheap "worker" model (with full repo access) executes code changes. We evaluate on 200 instances from SWE-bench Lite across five configurations that vary the manager-worker relations