I don't think that's true right now. LLMs are too unreliable, they will still need to learn enough code to verify the outputs. But even if eventually the code becomes abstracted away then they will learn other aspects of programming, like structured problem solving, repeatability, scalability, etc.