Hacker News Reader

I find that subagents usually burn more tokens and take longer, and produce about the same quality. A real killer use-case is using a VERY cheap subagent to do a lot of work, or reviews. Don't be fooled into thinking that a "scout" subagent will gather enough info for a "coder" agent to just start working.

View on HN

OMP also does a lot of rewinds - where the orchestrator comes up with findings on its own and rolls back to a previous turn with a summary update. I don't know if it always chooses correctly between rewinds and sub-agents, but it seems to do a pretty good job. Luna is SO MUCH cheaper that even if it spends 30K overhead in context its still much less costly, and when they work in parallel its faster too.

Reply on news.ycombinator.com