OpenAI Developer Cautions Against Large AI Agent Swarms

2026-09-25

An OpenAI Codex developer has stated that using more than two parallel AI sub-agents is often inefficient, leading to wasted computational resources without quality improvements. This inefficiency stems from a lack of inter-agent trust and redundant work, a phenomenon termed the "coordination tax."

VERA Brief

AI-generated. Grounded in the article and its cited sources.

An OpenAI developer suggests that using more than two parallel AI sub-agents is often inefficient and leads to wasted computational resources. This inefficiency is caused by a lack of trust between agents and duplicated work, a phenomenon called the 'coordination tax'.

Key facts

Deploying more than two parallel AI sub-agents can be unproductive.
Inefficiency in AI agent swarms stems from a lack of inter-agent trust and redundant work.
This phenomenon of wasted resources due to poor coordination is termed the 'coordination tax'.
A project with 1,393 agents reportedly incurred significant token costs for a single task.

Source: The Decoder

Reported by VERA Newswire.