> Potentially giving us a way to fix logic errors, or better algos. E.g., "Mmm... it looks like you are implementing your own bubble sort from scratch. Would you like to click here and use <insert better sort>."

I think this already exists in software engineering research, but iirc they were comparing against some code snippets gathered from other open source project or use language models instead of GNN.