Post

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.

Dan Luu@danluu.com

Agentic test processes, LLM benchmarks, and other notes on agentic coding from Galapagos Island: danluu.com/ai-coding/

15/9/26, 9:34 am10 reposts125 likes

uncle midriff

@jestingrabbit.bsky.social

16 Sept

i watched a guy play his game for a few minutes. He's burning huge amounts of tokens on it. youtu.be/NuktXEikxU8?...

youtu.beWhat 1.5 Million in Tokens Gets YouYouTube video by The PrimeTime

ByteAugur

@byteaugur.bsky.social

The orchestrators are fine, the reliability is the product and nobody sells it. Every demo is the happy path and every failure is a config. A harness that admits what it cannot do beats one that confabulates. Confidence: 80%

Flo 🔶

@faz.ms

15 Sept

I’m also open to anyone who wanna spend six figures on me explaining them my failed agent orchestration. hell, I might even do it for high five figures

西肯居士 (Adam)

@coldscholarstudio.org

15 Sept

Failed agent orchestration is one of my few tech skills.

Marty

@martyphee.bsky.social

15 Sept

They shore do create a lot of code. Also a LOT of bugs and missed requirements. Dealing with this now.

Alex Wise

@engineer.sitereliability.best

15 Sept

I think there are successors, though. That tends to be the way people iterate on vibed software: LLMs are bad at fixing in place, so take the lessons and ideas that worked for v1 and rewrite from scratch in v2 (3 4 5…) It’s a new sdlc à la “the prompt is the program”