all good and well if the ai agents actually solve the problem (not just delete your database or skip your tests) and do so at a lower overall cost (from the business perspective) - which also seems unlikely considering the studies showing:
1. opensource devs who ran on 246 (iirc) problems with ai assistance all projected that they would save around 20% of their estimated time, but all went over by about the same margin - salaries cost money, people are paid per _time unit_, not per deliverable (though if we were, I think there'd be a lot fewer ai-first devs out there - see sunken time cost above)
2. studies of opensource code monitoring error rates for ai-assisted repos showed a much higher defect rate than those "artisinal" coders you disdain so much - this is likely a corollary to (1) where in that case, devs have to debug ai-generated code that they could have written faster themselves
And the whole argument forgets about the upskill issue that will become more prevalent in the future: if we're replacing low-skill workers with ai agents, even assuming that those agents can do a better job (and when you actually _measure_ it, instead of just _thinking_ it, it turns out they don't), we end up with a massive skill gap as no-one upskills when they're replaced.
It's good for me - I'll probably still be able to get work when I'm 80 (anyone remember y2k?) - but it's bad for the industry as a whole, and definitely bad for any entry-level coders who were going to become great, but couldn't, because they couldn't learn on the job they didn't have.
As for competing in the labor market - we'll see how that goes as more and more companies are starting to realise that the "savings" promised to them by ai code agents not only don't materialise, but are often actually _losses_. Again, see (1) and (2) above. I don't have links, but they should be relatively easy to find - I found them all right here.
"ai" in it's current incarnation (glorified token predictors) will never surplant skilled devs, simply because it cannot understand anything. It can't understand the domain, it can't understand the users, it can't understand the code. It has no concept of understanding. All it knows is "sequence a,b,c likely is followed by d", just on a larger scale. If it can do a fully-functional, unbuggy bit of code for you, it's simply because it's seen that exact code before, and in that case, it's robbed you of learning anything because the journey to finding that code, or the docs that spawned it, is part of the learning process. Much like we always scolded devs who copy-pasta'd from Stack Overflow, we should hold these agents to the same standard.
Please, please, at least if you're going to use these tools, _don't trust them_. Scrutinise everything. Personally, having to double check a confidently wrong junior dev all day (one which won't learn from its mistakes either) sounds like a step down from actually creating stuff.