> Based on listening to engineers on various podcasts, almost all of them describe the current level of AI agents as being equivalent to a junior engineer: they're eager and think they know a lot but they also have a lot to learn. But we're getting closer to the point where a well-thought out Skill [1] can do a pretty convincing job of replacing a junior engineer.

The people that comment as such are either so disconnected from the software development process or so bought in on the hype that they are forgetting what the point of a junior role is in the first place.

If you hire a junior and they're exactly as capable as a junior 3 years later (about how far we're in now) many organizations would consider letting that employee go. The point of hiring a junior is that you get a (relative to the market) cheap investment with a long-term payoff. Within 1-2 years if they are any good, they will not be very junior any more (depending on domain, of course). There is no such promise or guarantee with AI, and employing an army of junior engineers that can't really "learn" is not a future I want to live in as a mid-career senior-ish person.

Of course, you can say "oh, it'll improve, don't worry" but I live in the present and I simply do not see that. I "employ" a bunch of crappy agents I have to constantly babysit only to output more work "units" I could before at the cost of some quality. If I had spent the money on a junior I would only have to babysit for the first little while and then they can be more autonomous. Even if they can improve beyond this, relying on the moat of "AI" provider companies to make this happen is not exactly comfortable either.