I think we can reduce the funny parts to what we think smart is by looking at the linear progresssion of llm itself.

prompting, can it reasonable get close to answering. context, can it understand my question and perform an action.

So the schools of thoughts should be folding on itself here and aligning.