One-line review after a month with Copilot
Writing boilerplate is definitely faster. The catch is that reviewing whether the code is actually correct takes more time. In the end, you still have to keep studying.
9 answers
Agreed, it's the same in real work. Generation takes 5 minutes, but review takes 30 lol
But is that secure? A lot of people are against putting internal code in it.
Well, that’s a bit of a stretch. Isn’t the longer review time less of a trap and more that you just haven’t gotten used to the tool yet? Once you’re familiar with it, reviews get faster too.
This is accurate lol — it’s not a real productivity boost; the barrier to entry just got lower, not that people’s skill levels went up.
Got a source? Is there a paper or something about this 'increased review time' thing?
I think just being faster at boilerplate is already enough of a benefit. There used to be plenty of times when I'd spend a whole day on that kind of grunt work. That said, as you pointed out, if verification costs are added, the net benefit will probably vary from case to case. I use it together with writing test code, and that combination seems to strike a relatively good balance.
It seems the fact that studying is ultimately the answer hasn’t changed.
Isn't that an old story? Models have gotten so much better now that review time has dropped a lot. I'm curious which model you tested with recently. Sharing the version and prompt settings would make it more relatable.
I felt a bit differently. Boilerplate is genuinely convenient, but that was exactly what scared me—it blurs the fundamentals. When I look at code I wrote a month ago, I often think, "Why did I write it this way?" Using it like autocomplete can make my thinking shut down. So these days I intentionally set aside days when I delete all the generated code and rewrite it by hand. It's slow, but I think that's what ultimately sticks.