I have explicit markdown about telling the model to not write comments. "Every time you consider writing a comment, instead consider re-writing the code that questioned you to write said comment to begin with. Write comments only when logic is complicated or unclear, otherwise 'comment' via naming."

The results thus far have been much better.

I'm seeing code reviews at my work where indeed, we have 10 line comment blocks for a line of code and now I just straight up don't read comments.

Sad state of affairs -- (emdash deliberately used here) but I guess the sooner the human gets out of the loop the better in this new world.

If you’d like to make dubious trades that’s your prerogative and who am I to stop you.

it's trivial to reimplement a better solution.

It's now trivial to fix these problems while still doing our day jobs -- shipping a product.

I run a team and am spending my time/tokens on serious pain points.

I presume I'm not the only one.

When you do this with an outsourced team, it can happen at most once per sprint, and with significant pushback, because there's a desire for them to get paid for their deliverable even if it's not what you wanted or suffers some other fundamental flaw.