About Copilot

Every now and then, someone asks me about Copilot. “Are you using it?". No. “Have you tried it?". Yes, of course. “You should use it, it makes people much more productive!". Well…

I have a vague concern on this topic, and I still have to put it into a precise set of thoughts.

First of all, my personal experience.
Personally, I found that using it was slowing me down. Because instead of just being in the flow, and focusing on what I want to accomplish, now I have to constantly check if the suggestion was good or not.
So, after a couple of days, I simply uninstalled it.

The problem I see, is that software development is not a repetitive task. That’s also the reason why estimates are so hard, by the way!

However, all the benchmarks we see about copilot are against well known problems, like interview questions. Moreover, every case that I see of people telling that it helps them a lot, it’s mostly about boilerplate code.

In other terms, it appears to me that those AI outputs are essentially copy&paste on steroids. Which is also reasonable, if you think that the basic mechanism is “predict next word from previous input”.

But again, this is not what I do. I usually try to remove repetitions and avoid boilerplate code. I try to work on problems that are not so common on the internet, where I can be a bit more creative.
In such context, I can’t see Copilot really of help – actually, it was even distracting.

Overall, I might be using in the future something like a better autocompletion, but I can’t imagine myself relying on such technology too much.
Maybe I am just too old school :)


musing

291 Words

2022-10-02 23:22 +0200