Usually more proficient engineers fix difficult to coin issues in large codebases, that require some notable traction with the codebase first.
Sometimes solution is a one liner, that's result of 1-2 hours of debugging. I can imagine AI helping with that only if it has in the context whole codebase, it's history, and good understanding of used libraries and languages.
Thanks for the article I also like the "scientific" approach.
Still, I have my doubts and I think some context is needed to put these numbers into perspective.
- Were these tasks something like coding katas? How well do they reflect actual problems that need to be solved by the developers during their "real" work? The work feels kind of "synthetic" if you know what I mean.
- Obviously writing the code is only part of dev work and figuring out how to write code that add value is a very different thing (that can' t be solved by current coding assistants)
- I find it really astonishing that an improvement of 40% only leads to moderate feedback about the perceived usefulness.
Again, I was very sceptical about coding assistants but tend to be a more positive now. So this is an very interesting data point - but probably exaggerating the potential overall productivity improvements considerably.
Question is what were the tasks?
Usually more proficient engineers fix difficult to coin issues in large codebases, that require some notable traction with the codebase first.
Sometimes solution is a one liner, that's result of 1-2 hours of debugging. I can imagine AI helping with that only if it has in the context whole codebase, it's history, and good understanding of used libraries and languages.
Thanks for the article I also like the "scientific" approach.
Still, I have my doubts and I think some context is needed to put these numbers into perspective.
- Were these tasks something like coding katas? How well do they reflect actual problems that need to be solved by the developers during their "real" work? The work feels kind of "synthetic" if you know what I mean.
- Obviously writing the code is only part of dev work and figuring out how to write code that add value is a very different thing (that can' t be solved by current coding assistants)
- I find it really astonishing that an improvement of 40% only leads to moderate feedback about the perceived usefulness.
Again, I was very sceptical about coding assistants but tend to be a more positive now. So this is an very interesting data point - but probably exaggerating the potential overall productivity improvements considerably.