Agent swarms waste tokens, says OpenAI dev
OpenAI Codex developer Eric Provencher warns that running more than two parallel sub-agents usually burns tokens without quality gains, calling it a 'coordination tax' and citing costly examples.