My experience is that "coding skills" are the intersection of analytical (being able to formulate hypothesis and test them) and language. There's a (small) barrier to entry. That and the lack of any gatekeeping is why we saw the explosion of Coding Bootcamps "become a javascript engineer in 12 weeks". Even tough the great majority of people attending these camps won't be able to write a single line of code at the end of the program, a few will be able to grasp it enough to fill in the blanks on a template for a website or an app.

We also had language models that could generate code out of very precise specifications back in the 90's. They were called offshored programmers. Send an email in the evening and get a .zip file in the morning with code. Sometimes it would even compile!

Engineering is much more than that. It's about being able to formulate, understand and analyze (conflicting) requirements as well as the ability to gain domain expertize as needed. The best example of that is Carmak and binary space partitioning. [0]

To me, what GPT is going to achieve with it's current performance is akin to what compiled and interpreted languages did: You no longer really care about word size, endianness or alignment issues. The compiler takes care of generating the correct assembly for you. Same thing with snippets generated with GPT.

[0] https://fabiensanglard.net/quakeSource/quakeSourceRendition....