Discussion
Vibe coding... then Vibe analytics.
Wanted to test out the expensive model driving cheap subagents pattern so put in a simple prompt for a web Tetris and set it loose. Kimi K3 driving Kimi K2.6.
Then dropped in the prompt Amplitude offers for a new project.
Then dropped in a vibe dashboard prompt into Amplitude.
The app is live: https://shermozle.github.io/tetris-web/
@shermozle how do you get an LLM to guide an LLM??? (and why?)
@decryption With the cheaper models it's all about keeping the context small. The expensive ones can hold a lot. So you get the big fat model to plan and instruct the smaller ones, then evaluate what they return and co-ordinate it. Most of the work gets done by the cheap ones.
@decryption OpenCode subagents, btw
@shermozle hmm, I'll have to play around more
Install bonfire.cafe
Get the full app experience
