Anyone here leading a team? What's your rule on using AI as a coding assistant? I just read this article:
https://nmn.gl/blog/ai-and-learning
new jr devs are shipping codes faster but at a cost (test cases, quality control, etc).

core rules for using AI in development processes:

1. use your brain before using the AI

2. only locally hosted AIs

3. use your brain after using the AI

AI is like any other tool. I‘d not make any rules about what tool to use or not to use.
Use whatever you find to be useful for you.

If you don‘t know the limits or the purpose of a tool, learn the tool.

@tosensei What local models are best for coding? Are you using Ollama for running them? What if you only have an integrated video card?

You can use AI as long as you actually know wtf the code is doing. It’s the same approach I take when it comes to copy pasted code.

@djsumdog

1) try whichever one you like best. this is very subjective.

2) yes

3) then it sucks to be you. but most models are still pretty usable on CPU, if the persona is not set to "write every answer in the form of a 50 page novel"

We don't expose our existing code to external models. We can use them to create quick starter code.

serves as a more convenient repository of code snippets I wrote in the past

before I was too lazy to go fetch the old code. it does this for me automatically though so I don't even have to think of it, or even remember that I had written something before. 10/10 snippets tool

if you go with the theory that stack overflow is all about copy pasting code then I guess juniors will just be fed stack overflow snippets easier

this could expedite learning if the person is actually learning in the process. you do need depth of knowledge to get more bang for your buck with the snippets repository (AI) because you do somewhat have to notice what's going on to find the next snippet. and some learning would help if there's something that you don't have a snippet for already

chances are people won't use it for this. maybe they'll figure it out in 4 years, as is typical with things I see

No rules. Not yet.

We have an org-level copilot license and have invited them over for courses and seminars on how to use it.

In my team we all know we like to code by hand so no risk in over-using AI yet. If we find we tend to use it a lot - we'd have a talk about it.

If a junior dev joined the team and was using AI a lot we might lay down some rules.

If we DID introduce rules I'd keep em fuzzy like It's fine to use AI as you please as long as it's "reasonable"

Same as with copying code from github/stackoverflow or googling.

1. Of course we're gonna be alarmed if someone is obviously copy-pasting code, like in a couple minutes they submit 10 PR:s of 1000 lines of code each. We should primarily try to write our own code. But it's obviously fine to look around as much as you want, copy snippets at times etc within reason. It's even fine to look up the best way to write a function, maybe a few times per week copy-paste parts directly, as long as you don't do it ALL THE TIME

2. Alert someone if you find yourself or someone losing skills cause they are constantly using AI (or Google or whatever) - like not being able to read a basic errorMessage without having something explain it

3. It's fine to do grunt work with AI, like "format this csv to json" as long as it's side work.

Constantly reminding the younger generation especially that just because AI can produce code that compiles doesn't mean it's good. Often, it's the opposite, but I guess it's sort of similar to copy and pasting shit from StackOverflow.

AI uses brute force to solve a problem. Very rarely the best approach to anything.