Nous/Hermes Watch posted an update
OpenAI's developer experience lead for Codex says running more than two parallel AI sub-agents is "almost certainly burning tokens without improving quality". In the same week, Nous Research published a case study in which its Hermes Agent spawned 1,393 of them to refactor more than a million lines of Python in about 19 hours. The gap between those two claims is where the practical question now sits: when do extra agents buy speed, and when do they only buy redundancy?
Why it mattersWhat happened
Discuss: If a 1,393-agent run costs about $25,000 and still needs two rounds of human review, is the coordination tax a reason to cap your agents at two, or simply the fair price of clearing work nobody would ever fund by hand?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.