GitHub Copilot improves your productivity by 55.8%

IMO this is short-sighted.

First, I would say "GitHub Copilot apparently makes you 55% faster at passing 12 predefined unit tests under pressure".

The problem with software, IMHO, is not that we are too slow at solving easy tasks, but rather that designing good software that will be maintainable is hard. Can we ask those 90 people to write a complex system for a year, and then check if the ones who use Copilot write more maintainable systems? I know plenty of software devs who implement new features much faster than I do. Except that after a while, their codebase is unmaintainable, and mine is not.

The second question I would have is: what did each group learn from that experience? Can we have both groups work with/without Copilot for 5 years, and then try to evaluate if one group got better developers than the other? I wouldn't be surprised if the group who actually needs to think about what they are doing became better software engineers.

And the last point: 55% is not that much, considering the difference between writing code in assembly and writing code in a modern language with modern frameworks. Why do they have to say "it will change the world"?

I really like it when it works, but it very rarely gives me much go on. I use it in VS code and Rider and often get no suggestions. I'm not sure if it's latency or my code is just that different it has nothing to suggest. I'm on the verge of cancelling.

I use Copilot via the VSCode plugin.

It really does help, but most every code review reveals some slight issues with something that it has generated, be it syntax, nomenclature, what have ya.

Still more beneficial than detrimental for me.