I mean, isn't this with achievable prompt refinement?
One could argue that it wouldn't satisfy some $arbitrary_dimension (scalability, accessibility, whatever), but you can imagine it wouldn't be hard to get there with prompt refinement eventually, especially on an AI trained to satisfy SWE requirement (and not talking about GH's Copilot,which is an early application of a fairly young technology)
What would the co-pilot equivalent be? You give it a prompt, it generates a few repos, but then you still have to understand the whole project to judge that it respects the system needs and restrictions. After a few rounds, maybe you will have to change up the code a little, but 90% code will be written by the machine. I, for one, welcome this future -- because once again, this "prompt refinement" iteration is where our human creativity is used most effectively; I think it's a significant improvement over writing boilerplate code every time.
I'd be more than happy to get rid of the "programmers" that are generating the same quality code as Copilot. Engineers are employed to translate human needs into tangible system designs, that they have to write code to do so is mostly an artefact of our current system-building methodology. In fact, I would be immensely happy if I could get rid of that part.