I don't think this line of reasoning is very reasonable anymore. LLMs are getting good not only at producing high quality output but even preemptively dealing with architecture and other 'big picture' type designs. And this has been true for quite some time, certainly at least since the Fable gen of LLMs.

And while I do expect progress to eventually slow, there's no reason to yet think we're at the plateau point or necessarily even especially close, so they're probably only going to get better. But at this point, even if they didn't I'd still be saying this. 'Programming' with an LLM is a loathsome and unpleasant task, but that's because it's no longer programming but instead an entirely different design-centric domain where it's handy to have some programming knowledge to nudge it in the right direction sometimes.

I think human programmers are going to go the way of human computers. [1] And I'm obligated to link to that because now most people probably no longer even know such an occupation existed, even though it rather obviously must have if you think about the past for a moment.

[1] - https://en.wikipedia.org/wiki/Computer_(occupation)

The quantity over quality argument has substantially broken down in the last 6 months especially. SOTA models outperform the vast majority of engineers at engineering tasks, let alone everything else they're also exceptional at, including system design, incident recovery, DevOps, etc.

This is because these models act as a productivity enhancer for engineers. A 1x engineer becomes 5x. A 10x becomes 50x. And so on.

Slop is a risk that can be mitigated through team policy, skills and other mechanisms. Nothing we produce is slop, in fact incidents are down dramatically. It's more stable than ever.

Nobody is measuring LOCs. They're up, but that's barely relevant to me. I'd rather measure defect rate, rollback rate, sprint velocity, roadmap adherence etc. Those measure actual progress.

I don't use LOCs as a reason to manage people out. But if they're vehemently refusing to even try AI, other metrics will make it obvious that they're falling behind.