@cammerman I'm seeing the writing on the wall with this currently. A higher-up has been using Claude extensively to create an entire, complex project. They do know how to code well, relatively-speaking, but admit that they don't even read the code anymore. Instead they've built an elaborate scaffold of processes for Claude to follow, and extensive testing. It's impressive, but also frustratingly wasteful and short-sighted.
Part of me wants it to blow up spectacularly. The annoying thing is...if I were asked to contribute when this project was being started, I probably would have really enjoyed designing things. Instead, now it's a big black box of AI-generated code that is hard to understand and make meaningful contributions to...unless you also use Claude, which I have, and it was miserable.
So to your point -- when software devs actually *know* a codebase, there's zero friction to making changes. Anthropic is basically selling a solution to a problem they encourage people to make.