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

Eric Provencher, a former OpenAI Codex developer, suggests that deploying large swarms of AI agents is largely inefficient and yields minimal performance improvements. He identifies a coordination tax where parallel agents spend excessive tokens redundantly verifying each other's work rather than increasing task accuracy. This finding challenges the current industry trend of scaling agentic workflows by simply increasing the number of active participants.