LLMs are fairly good at understanding how a bunch of naming in slightly different forms maps around a system. This is basically the one trick that transformers/

LLMs are fairly good at understanding how a bunch of naming in slightly different forms maps around a system. This is basically the one trick that transformers/attention does! I think a smart copy/paster is what LLMs will be good at. What they won't be good at is when there are deeper semantics, when there are performance requirements, when there's a complex data model, or when you need to come up with the data model and consider how that will work over time.