Post

hailey
@hailey.at

i do think it’s very funny whenever something like this gets portrayed as “revolutionary and new”. this has been a thing you can do for nearly half the year now, just none of the publicly available harnesses allow for it and people trying to do it get called “insane” unless they work at a lab

mr. TIM@timkellogg.me

OpenAI is deep into developing RSI & RL’ing multiagent systems here Noam Brown describes a multiagent system that doesn’t do top-down management

hailey
@hailey.at

if this is a “most recent moment” for him, he is wildly behind the times of what people are building lmao

Jake Gold
@jacob.gold

Sure, you can do it, is anyone doing anything like this for real work? Haven't seen it in any of the startups I've visited. Claude/Codex harnesses can spawn lots of subagents now, which is useful, but they use the manager-worker model because that does work well. Things like Gas Town were hot air:

Dan Luu@danluu.com

Interesting to see Steve Yegge say he never successfully built anything with Gas Town. I mentioned not finding these super vibed orchestrators useful because the reliability was too low (w.r.t. completing non-trivial tasks). Turns out the author of the most famous one had the exact same issue.

Carl Vitullo
@vcarl.com

Extremely anecdotal, not actual work, but I had four independent Sonnet agents playing an MMO with access to self-modify the SDK that hit the game's API. They used in-game chat to discuss bugs they noticed and notify each other of commits they pushed, and a bunch of them were real fixes

1 repost3 likes

Carl Vitullo
@vcarl.com

I shut them down when the someone in the in-game forum started recruiting for a cult and my agents burned my weekly limit theorycrafting someone's fake religion