I have been really impressed with GitHub Copilot for Roc while using it for AoC this month. It more often than not makes helpful suggestions. It is especially useful for generating type signatures. Sometimes it gets the syntax or standard lib names wrong, but those are easy enough to correct.
My only other experience with Copilot is in large Ruby on Rails applications which I have found it less helpful. This surprised me because Roc is such a new language and has less training data. Perhaps it’s because Roc is a typed language, or that Copilot is really good at AoC puzzles?
What’s your experience with Copilot for Roc compared to other languages?
Or that the quality of ML code on the web is better than the average Ruby code. Or that the structure of ML code is less ambiguous than Ruby code.
it could also be partly because the kind of problems that copilot is solving for Roc and advent of code are different to the kinds of problems you're solving with ruby and rails. :thinking:
Last updated: Jul 05 2025 at 12:14 UTC