Is it actually worth having GPT review your code?

Lately I've been running my code through GPT before committing, and it catches more than I expected—things like null handling and edge cases.

But sometimes it points out things so trivial that it feels like a waste of time... Where do you all draw the line?

by 무한도전러482

6 answers

Agreed, you really nailed the null handling lol. I caught two NPEs myself recently.

by 스타트업러411 · ▲0

I only run it on a per-diff basis. If I feed it the whole file, it nitpicks every trivial detail… When I give it just the changes, it points out only what’s genuinely necessary.

by 클라우드러버338 · ▲0

Well, if you trust that and merge it as is, you'll run into bigger problems later. Ultimately, a human has to make the final call, so use it only as an assistive tool.

by 주말개발자137 · ▲0

Is there a source? It probably depends on which model it’s run with.

by 알고리즘고수788 · ▲0

But isn't that an old story? Isn't that something modern IDE linters or static analysis tools already cover well enough? Most of what GPT catches is stuff tools were already catching—I don't see why you'd burn tokens on it. And if the review feedback is trivial, I think that ultimately means the prompt wasn't written well. If you limit the scope like "only focus on security/performance issues," it becomes much more useful.

by 취준생김씨677 · ▲0

If I get vague feedback, I just ignore it lol. I think it’s better to have clear criteria. I only apply security, performance, and obvious bug issues, and toss style comments. After using it that way for half a year, I’ve felt code review time actually decrease. That said, when GPT says “this looks like a bug,” sometimes it’s a real bug, but there are quite a few cases where it insists a bug exists when it doesn’t, so I banned just copy-pasting and applying its suggestions. Once it flagged a perfectly fine null check, so I asked why, and it turned out it had misunderstood the logic itself. In the end, I only use it for reference.

by 프롬프트장인579 · ▲0