I ask the AI to monitor this on its own; my agents have automatic reminders once a week - to check all the code for technical debt, suggest solutions, and sort tasks by risk and complexity. And, in fact, this helps me not to overlook anything.

My style of working with AI is...

"I'd like to do X but I am concerned about Y, and Z, what do you think about that?"

... discussion ...

"Ok, I'd like to ask you how you would go about doing X but before we do that do you have any questions for me?" (with 'somewhat' detailed requirements, like a paragraph or two)

... discussion ...

"So how would you do X given that?"

"Ok, I don't understand about A, explain it to me better"

"Are you sure about B?"

"My supervisor says we can't do C, what else can we do?"

"Make it so!"

"hmmm... that's pretty good but I don't like D, can you do E instead?"

"so, that's really good but when I tested it I found this little problem, can we fix it?"

"I've never seen the syntax you're using at point F, how does that work?"

... now it works and I make a commit ...

"Hey, I made some edits to the code to make it my own [1]. What you did think? I didn't make any mistakes did I?"

... delete excessive comments, maybe add a comment, do some more testing, commit, push, close out ticket ...

[1] ... be consistent with my style, be more clear

10% coding 90% testing and fixing issues with the code.

It might not even be as much as 10%

Just as previously, crash and burn.