“AI” has always been about automated programming even before it could generate code. For any problem there exists solution program. Coders would design imperfect programs and then trial-and-error (search) until it works. Neural networks arrive at solutions using only search. But you still needed classical code because neural networks used to be difficult to train and use. LLMs are neural nets that went meta and the problem they set out to solve is language, of which programming is a subset. So now neural nets are fully accessible and “trainable” in English. Where you need code today is the program skeleton and the glue code orchestrating and moving data around—which LLMs can help you write! So today anyone willing to learn a little bit of coding and can think clearly can make pretty powerful software.