How do you know if you're being overly dependent on AI or you're just tired/lazy to debug everything?

How do you know when to stop using AI for debugging errors (not becoming overly dependent on it)?

I would catch that across multiple foundation models and AI tools that they hallucinate and invent an npm module that doesn't exist or a Swift framework that doesn't even exist!

Sometimes though, it would be a godsend and save me from spending hours of time digging through documentation on the tiniest minute details that almost no one would catch.

I'm trying to strike a balance, since I mostly 'raw dog it' as some of my friends at Taro say. I normally write my own code, but I've been using this to debug a lot of boilerplate code from existing open source frameworks.

I make sure I understand every single line it is generating for me (and it works most of the time), since some of it I also use to edit tiny syntax.

I will still use cli/terminal/ViM and write out all of my other installations (muscle memory for engineers) and write some starter code and not give away all my power to an AI.

AI for mehas normally I use as a replacement for Stack Overflow to chat with me to do research on what I'm debugging to try different things if I get stuck, or to reconfirm small syntax errors 'like hey you missed a curly brace, a trailing comma' or some tiny error somewhere.

Overworking

Sometimes I find myself overworking, so I am trying to find more natural stop points or give some of the minute/minor workload to AI so I don't overdo it, but I find if I surrender TOO MUCH control to the AI, then I'm prone to way more errors.

What's a good gold standard to strike a balance for using AI tools for software engineering?

I know and lots of my friends due to its ability to hallucinate, even if can make you more productive, I can't expect it to do the heavy lifting for me, and only really generate boilerplate code I could write myself that saves me time, and be my Stackoverflow replacement.

In the end, I still have to read and write code (and evaluate what the AI is doing which). Yes sometimes, this can take up more time and waste time. Other times though it can save me hours of time on harder problems that are less well documented online and give me solutions that are actual helpful).