LLMs will for the forseeable future become amazing at simple tasks. Everyone is trying to make them do difficult tasks, which I think is not the right goal, at least for now. Being able to do extremely simple tasks together with a general understanding of language allows one to make different systems talk to eachother without needing a human in the middle. I think this is a good example.
I experienced the same thing with a small PHP project. After some iterations it just completely changed course and wrote the project completely differently.