How far do you all let GPT handle code reviews?
These days I just throw everything in before committing, but honestly it's half nitpicking and half helpful lol
But I get a bit hesitant about uploading company code, so lately I've been holding back. Do you all have internal policies for this?
9 answers
I'm with you. When it comes to company code, it ultimately comes down to accountability, so it's something you have to be cautious about—but if you document your decision criteria, it gets much easier.
Uploading company code is really crossing the line. I once saw someone get caught for that and have to go through training with the security team, and it was absolute hell lol. Just run that kind of thing only in personal toy projects or side projects.
facts
Well, isn't that a bit overkill? There are already plenty of models you can run locally, so why go that far?
Does your company have a policy? If not, I think you should make one. Our team set guidelines to mask sensitive information and key values, and to only use internally approved models. Just throwing stuff in—when you think about how the code itself gets trained, isn’t that kind of scary?
Got a source? Can I get a link to the official docs saying it was included in the training data?
I also run it before commits out of habit—half nagging, half helpful, I can relate lol. But if I just fix things exactly the way it tells me, it feels like it’s not my code but its code, so these days I only take the suggestions as reference and rewrite it myself. It’s definitely convenient how it catches little things that are awkward to ask a reviewer about (naming, early returns).
I disagree. Code review is ultimately done by people who know the context, and GPT doesn't know that context. Aren't most of the people saying it's helpful just getting called out for things that linters would catch anyway?
At our company, they just hook us up with an in-house model lol. We can’t use external APIs because they’re subject to data export review. I’d check and ask the infra team—that’ll probably be faster.