[Sparkyte]

Why not just write many small models for explicit tasks than running one bigger model anyway? I prefer the agentic subject matter expert design anyway. I suppose because it wants to look at the whole code base?