Well, that and debugging.

I was ok with the thesis of this article, right up to "letting AI decide". "AI" in this article is "LLM". LLMs don't "decide" per se. They give you something like the words other people wrote when they had a syntactically similar decision to make.

Using "Letting AI decide" sweeps that under the rug. If you're good with how basically everyone else decided for a syntactically similar problem, then fine. If you want something different, now you've got 2 problems.