AI agent swarms are a massive waste of tokens with zero quality gain, says OpenAI Codex developer

Eric Provencher, a developer at OpenAI Codex, warns that running more than two parallel sub-agents often leads to unnecessary token consumption without improving quality. He highlighted a project where 1,393 agents spent $20,000 in tokens on a Python refactoring task that a single Astra agent could have completed for a fraction of the cost. This phenomenon is termed the "coordination tax," indicating that agents do not trust each other and end up double-checking each other's work, resulting in resource wastage.