> and diminished sense of ownership over their own writing.

Anecdotally, this is how I felt when I tried out AI agents to help me write code (vibe coding). I always review the code and I ask it to break it down into smaller steps but because I didn't actually write and think of the code myself, I don't have it all in my brain. Sure, yes I can spend a lot of time really going through it and building my mental model but it's not the same (for me).

But this is also how I felt when I managed a small team once. When you start to manage more and code less, you have to let go of the fact that you have more intimate knowledge of the codebase and place that trust in your team. But at least you have a team of humans.

AI agentic coding is like shifting your job from developer to manager. Like the article that was posted yesterday said: 'treating AI like a "junior developer who doesn't learn"' [1,2].

One good thing I like about AI is that it's forcing people to write more documentation. No more complaining about that.

1. https://www.sanity.io/blog/first-attempt-will-be-95-garbage

2. https://news.ycombinator.com/item?id=45107962

Yeah, same experience here too. I "vibe coded" a project, about 3k loc including tests. But whenever I need to look at it for bugs etc it just feels like I'm looking at someone else's code. I don't have that intuition of where things are, which bits are a bit fragile, which bits might be the likely cause of an issue etc.

I mean, ultimately, I didn't write it myself. It's more of a "remix" of other people's code. Or like if I translated this comment into French. It wouldn't improve my French so why would vibe coding be expected to improve one's programming ability?