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

Eric Provencher, a Codex developer at OpenAI, is warning against reckless token spending in agentic workflows. In posts on X, he argues that more than two parallel sub-agents almost always burn tokens without improving quality. Agents don't trust each other, so they end up "double-checking everyone's homework."

Provencher calls this the "coordination tax" and says it's "really hard to keep so many parallel lanes running without spinning out and burning tokens on excessive verification." When a project surfaced where someone spent $20,000 in tokens refactoring a single Python file with 1,393 Fable agents, he commented that a single Astra agent could have done it for a "fraction of the cost." Agent swarms may save time, but the token overhead is a "trap," he writes.

As a fix, he recommends delegating tasks to separate threads that only notify the main agent once done, instead of constantly polling for status. System prompts also get executed multiple times across sub-agents, and sub-agents often lack needed context, causing redundant work. Provencher acknowledges OpenAI still needs to ship better solutions here.

AI News Without the Hype – Curated by Humans

Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.