I kinda agree with your first point, but AI assistance is so incredibly powerful that it's foolish not to use it, unless you're working on some really important logic. And even then, having an AI skim for common mistakes, inaccuracies or inefficiencies is still very valuable.

And what you're describing is really "Unix philosophy" and I strongly agree with that. Make a piece of software that does its one thing really well, and have it communicate with a simple API (POSIX).

In Unix/Linux you generally just "pipe" one program's output into another program's input, and can chain them virtually infinitely.