AI coding assistants (GitHub Copilot, Codium, etc.) are great. They help us spend less time writing boilerplate code, which has always been annoying.

What worries me though is that they will make us more accepting of the boilerplate code, despite not helping with reading it. And personally, I read code a lot more than I write it, so the equation would be rather unfavorable to me.

Designing efficient, concise, expressive APIs is a lot of effort, and AI assistants nullify the short-term payoff of doing so (i.e. you won't have to write verbose code). And humans notoriously undervalue long-term payoffs.