I think there exists room for an intermediate layer, a framework for AI to write code in specifically. With larger "target" code blocks it can stitch together.

I think there exists room for an intermediate layer, a framework for AI to write code in specifically. With larger "target" code blocks it can stitch together. Code is designed around humans to allow them to write specific, abritrary and precise logic with a lot of minuate, but LLMs might be better off thinking in larger business logic blocks that are testable and "sanctioned" by the framework makers for the specific problem domain.