Ask HN: 4x more output realistic with Copilot?

1 point·theyinwhy··4 comments

Hi HN, our managers (large multi-national enterprise) introduced Copilot last week and told us they expect developers to ship 4 times more features from now on. Is this a realistic goal in any way or form?

View on HN

No. I ran across the metrics in my company on Copilot usage. Something like 20% of suggestions were accepted. And in my own usage, I can tell you that rarely are those suggestions just accepted in a way where I can move on. I often accept it as a reference, and then end up writing my own thing. AI has not really helped me write faster code, it has mostly speed up searching for syntax or the right library to use.

Writing code is also not the bottleneck in my work. It's the meetings and getting requirements needed to write the code, or waiting for decisions to be made on what we're doing. AI isn't speeding up those things at all, so I haven't really seen any impact on speed of delivery.

Of course not. While some development acceleration might be expected, at best you'll get some multiple that is representative of the overall quality and experience of the devs you work with.

Copilot has definitely helped me speed up via not having to type a lot of boilerplate and line completion, but I'd say that would be at most ~50% speed up when it comes to greenfield development and half that when it comes to maintenance work (and even these figures are probably generous). A 4x speed up sounds ridiculous; expectations like this are how your codebases get filled with GPT generated garbage that is connected together with GPT generated glue and then wrapped with GPT duct tape and a bow

Amdahl's Law comes to mind. It can't speed up all the parts of the task, so there's a limit to how much it can speed up the whole.

Ask them if Copilot is making them 4x as fast at figuring out what the right features are.

Reply on news.ycombinator.com