A key developer behind OpenAI Codex has voiced strong skepticism regarding "AI agent swarms"—a popular methodology in current AI development that involves coordinating multiple AI agents. The developer described this architecture as a "dramatic waste of tokens" and criticized its contribution to quality improvement as extremely limited.
AI agent swarms have been gaining traction as an approach to divide and conquer complex tasks. However, critics point out that the mechanisms by which these numerous AI agents interact significantly increase the overall computational cost of the system without directly translating to improved output quality. This highlights a growing imbalance between excessive resource consumption and the quality of the resulting deliverables.
This critique serves as a warning against the current trend in AI development of over-relying on "complexity" and "the number of agents." It raises an important discussion on the need to move away from designs that simply stack agents on top of each other, and instead return to fundamental designs that optimize the reasoning capabilities of single models and maximize token efficiency.