[NotGMan]

The problem is in the very act of creating a false dichotomy of "vibe coder" vs "crafstman": the two extremes exist only in marketing/stupid blogger mentality.

The future programmer will have to use both until true AGI is created.

You cannot prompt yourself out of difficult scenarious without understanding the underlying mechanisms of the problem/solution.

The trivial prompting as in "it's not working for test X, fix it, it should be Y" work right now well only in more simplistic domains and cherry picked example.

This immediately becomes obvious in many gamedev scenarious where AI totaly fails and starts to produce garbage for even trivial things such as not being able to fix "Mouse Y should pitch camera up, not roll it".

I personally did found modern AI useful for some very obscure API cases for which there aren't many docs online. It didn't give me the correct solution but merely creating non-working boiler-plate was enough to give me the insight for how to fill in the blanks.

[brazzy]

The real question is: how will programmers who start out learning to program with AI doing all the straightforward, clearly delineated and specified tasks for them, be able to make the jump to steering and overseeing AI when it struggles with more complex tasks?

[CarlosBaquero]

Exactly, that is the one million dollar question.

[ncr100]

I don't disagree.

And, I'm skeptical agi will usher in human readable, maintainable, and auditable agent-built code. I suppose we'll have to kindly ask the AGI for auditable code.

[cma]

It's already good at making human written code human readable, not yet refactoring huge complex stuff but helping you read and navigate it.