Pre-AI juniors were valuable because on many tasks, it was faster for me (senior eng) to write a quick spec have a conversation with a junior, and have them go off for a few hours or a few days and write the code, and then come back with it for review. I could do that with several junior engineers and pretty reasonably paralellize the work.

Now a junior takes my spec, drops it into Claude, and submits a PR a few minutes later. So I'm back to being the bottleneck -- there's constant pressure to provide specs and review code, and ultimately that process is just me having an indirect conversation with Claude (em dash is mine).

The junior is probably providing negative value, since it would be more efficient for me just to talk to Claude, and they're not learning anything, because it's really hard to learn anything by skimming code that's being pooped out by Claude.

I'm not sure what the solution to this is, I still think we need to train junior engineers. I think my best advice to juniors right now is stop using Claude so much. Use it to plan and answer questions, but you should still be writing code even it's slower. Because that's the only way they're going to learn enough to effectively guide AI and review output and get past being a net negative.

Do you think that its possible to raise the bar for junior engineers such that "you have AI now, so I need you to own farther up the review chain yourself"? I feel like I could answer either way. No, you need reps on projects to have that kind of judgement, that's what makes a senior. But also yes, while it will take the junior way longer to do that kind of review, they have the time to do it now, and its now the bottleneck so it makes sense to put their effort into that portion.

> because it's really hard to learn anything by skimming code that's being pooped out by Claude.

I think this is pretty dubious. It's true that you have to do more proactive learning with an agentic workflow, but it doesn't mean you can't learn stuff. Software engineers forged in the age of AI are not learning the same things their predecessors did or thinking about code in the same way, but the more curious people will still be better at the job, which is I think how it's always been. It seems a little unclear how much "AI brain rot" will be a force unto itself that neuters even the smartest of the juniors, but I think it's still an open question.

This, pretty much exactly what I've been noticing too. Feel for the juniors though, its rough trying to learn when the tool in front of you can just spit the whole answer out. I still think they're worth it, but tbh I just don't have a good answer yet for how to actually use them well, maybe just takes time to figure that out.

The other challenge with real people is the emotional aspect. Easier and faster to tell ai the code it wrote is terrible and needs to be redone completely, or maybe you just end the session and use a new model. Trickier to do with a human.

SAT for junior developers; questions are code snippets and architectural decisions.