AI agent swarms are a massive waste of tokens with zero quality gain, says OpenAI Codex developer
OpenAI Codex developer Eric Provencher warns that running more than two parallel AI sub-agents burns tokens without improving output quality.
Eric Provencher, a Codex developer at OpenAI, said on X that parallel agent swarms incur a 'coordination tax' because agents redundantly verify each other's work. He cited a project where 1,393 Fable agents spent $20,000 in tokens refactoring a single Python file, arguing a single Astra agent could have done it for a fraction of the cost. He recommends delegating tasks to threads that notify the main agent on completion rather than constant polling, and acknowledged OpenAI still needs better solutions.
- Provencher says over two parallel sub-agents yield zero quality gain
- Cited case: 1,393 Fable agents spent $20,000 refactoring one Python file
- Recommends completion-notification delegation over status polling
- System prompts execute redundantly across sub-agents lacking context
From the original article
1 min read at the-decoder.com
OpenAI Codex developer Eric Provencher warns that running more than two parallel sub-agents almost always burns tokens without improving quality because agents don't trust each other and end up double-checking everyone's work.
By Matthias Bastian · the-decoder.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.