Yeah, and the thing is, we're nearly at the natural language stage for computers to 'understand' what we're describing. On the other end of things, compilers are creating machine language that's so far removed from our 'code', that you couldn't recognize your compiled code anyway.

So, what 'programming' is, in 2022, is using a specialized language to describe to a compiler what machine code to produce.

If you just had a little machine learning on both ends of that, things would change dramatically. The 'code' you describe could be less precise (no more missing semicolon), and the machine code produced might be much more compact and efficient (recognizing specific cases of matrix multiplication that can be optimized).

We're already basically doing this, it's just one little step to say 'why can't I just tell the computer I need to read through all the records and get the one with the user ID I'm working with?'

With co-pilot you basically can say that, and as that improves there won't be a need for 'programming languages', just natural language descriptions of what each piece does, and then as AI gets better, those pieces will get bigger and bigger until you're just describing applications at a high level.

Eventually you won't even have 'applications', you'll just describe to the AI what you need, and it'll provide it in something it knows is intuitive to you.