Quote:
Originally Posted by Shazam
Here's the problem though: You have to be a very, very, very good programmer to know when it is wrong. There is a mid-level developer (I wouldn't even call him that) on my project, he uses AI all the time, and he always trusts the answers. His code is a total ####ing mess. Often times it doesn't do what they ask, but he can't figure it out, so how the hell would he ever know if the AI did it properly? His part of the project has been on bug fixing for weeks now.
|
This is the problem I think.. if you start to really depend on AI for code how do you get good senior programmers? Good programmers come from the struggle and fight with code learning how to do things well.
But in my use Copilot has actually been pretty decent, but most of my stuff is pretty narrow scope and small functionality, and I find it most useful at code completion.. I'll write a line, Copilot will suggest a few more and a lot of the time it's right or close enough. But I know enough to know it's right.
It's also creepily good at comments.
But my boss didn't want to bug me he wanted to make a bash script to get info from a bunch of calls to a URL and parse out one piece. Didn't work.. or it worked but the info he was wanting was all empty.
I looked at it and it looked fine to me.. actually took me 5 minutes of debugging to finally figure out the source file he had made with the URLs to use was created in Windows but the script was in Linux so the CRLF vs LF characters were causing problems.
I'm sure AI could have fixed it to compensate, but how would you even know to ask for that kind of modification without being able to debug it and having the experience on how to debug or even know about Windows vs Linux stuff?
I don't think AI saves me a ton of time, but it does grease the wheels, reduces friction a bit while coding.