> Getting the initial code written is not the big deal. For curl, maintaining and polishing the landed code through decades is the real task.

This is the real sticking point for me with vibecoding. In the short term, vibecoding everything might get you something that works. But I have a hard time accepting that in the long-term, maintaining a large and complex codebase is feasible when you haven't actually developed a deep understanding of it.

But maybe that's my bias as someone who considers their job description to be basically "develop a deep understanding of and maintain this large codebase".