I read it and I wish I could agree with the points in the post, but I can’t. I love coding, I love typing my solutions by hand, but I also switched couple of months ago to AI generated code. If you do that in a targeted way, where it doesn’t have to build a "world model" of your codebase then it writes much better code than an average human being would. My codebase at work is about 220k lines of Python, Rust, C++ and some Bash. I can tell for Python it’s the best, for Rust, compiler helps a lot (some sneaky logic errors are still there though), for C++ it’s horrible (bloated sheeeet) and for Bash (well who cares about Bash ). So, what I am trying to say is, that it matters how you work with your setup and what the volumes are and if you can still keep up with proper reviews or not. I see literally two ways going forward, either full control and tiny gradual, marginal changes (like before AI) or full vibe-coding. There is nothing in between.

> much better code than an average human being would.

Assuming you mean an average software developer, that's still an incredibly low bar.