New GitHub Copilot Research Finds “Downward Pressure on Code Quality”
I’ve noted down some thoughts about research that shows CoPilot use, lowers the quality of code, overall.
I don’t find this surprising, but I also don’t think this is a damning indictment of tools like CoPilot. They can be useful to unblock you if you’re stuck, but I sure as hell wouldn’t trust the code it suggested verbatim.
This quote is interesting to me:
Developers who used GitHub Copilot completed the task significantly faster — 55 percent faster than the developers who didn't use GitHub Copilot.
Do I think that’s a good thing? No, not really. I’ve always been averse to “we need to do it faster”, mainly because a need for speed can quite often create mountains of tech debt and burned out developers in the long run.
I always say move slow to go fast. What I mean by that is taking your time to do as much pre-production work as possible — carefully considering the impact of your code — might make things feel very slow, but the reward is everything gets faster in the long term, because shock-horror: planning makes things simpler, usually.
So yeh, I don’t think stuff like CoPilot is net-bad. I — as always — find that when it and other tools are leaned on to speed things up: very rarely will these be good long term solutions.