[sph]

We’ve invented a machine that can spew a lot of garbage out. Sure, you can sift through it to find the nuggets of gold, but the proponents of this new methodology heavily discount this additional effort. Studies seems to show that at the end of the day the benefits are minimal, despite the claims.

Once again, it is a sign of modern programming that the solution to more problems is throwing more code at it, rather than more upfront thinking that will lead to less code overall. The “work hard, not smart” crowd won this round, I guess.

The best engineer used to be the lazy one [1]. The mental effort of micromanaging a machine by explaining what you need in prose is diametrically opposite to just being lazy, sitting on a hammock, and writing directly the simplest possible solution that requires the least amount of maintenance. But sure, enjoy the 200k line vibe coded ball of mud to run ferrets and gas towns or whatever these guys do all day.

1: before dopaminergic stimulants were as widespread as today.

[assaddayinh]

Its a opportunity, all these ai first companies will die in 5 years from tech debt, if you can human engineer a good replacement thats ready to take over at that point in time.

[jaapz]

It doesn't have to be black and white. I use the agent just as another tool in my programming tool belt. It shines especially in the boring repetetive tasks that would otherwise not be done because I can't justify the needed time. The choices arent just "no agentic AI" and "vibe coding only", there is a middle ground.

[powersaustin]

I think the issue is that there is a lot of work that is low value but not low risk. It has always been danger to give that to juniors (or have too many juniors) but it at least trains them. If AI adds more juniors then you are basically making a bad resource setup along the lines of the interns are free model.