I'd be very cautious about calling LLM output "an abstraction layer" in software.