In A.I., we see this with hard-coded, assembly FSM’s vs interpreters running high-level code. The former works with high-efficiency out of the box but can’t change behavior or improve much. The latter does nothing until it’s taught the extra knowledge (interpreted codes) which also might have new behaviors (functions). Many game developers switched from hard-coded assembly to interpreted code for AI agents for that reason.
So, it’s not under-developed: it’s a better-developed component with different tradeoffs.