I guess hardware being able to run a local model will eventually get cheap enough, but for a lot of people even buying an Apple device or something with a good enough GPU is prohibitive.

True, it will get cheap to run today's frontier models. But, by that time, how much more advanced will the frontier models of that time be.

It is a real question. It all depends on whether the AI future is linear or exponential.

I think we are already there. You can run a pretty ok LLM on a 4gb raspberry pi that will write most any simple 20-150 line bash script today, or toy application in python/rust. Old laptops pulled out of the trash are probably capable of running smaller LLMs and can explain how functions work. They're no claude code but you probably want a rough-around-the-edges LLM that can't do everything for you, if you're planning on using it to learn to code.