We need to talk about migrations with AI

I think that this is one of the better applications for LLMs. Their ability to translate text from one format to another is one of their strong suites. Especially with things like testing libraries, there is immediate feedback on whether the translation was correct.

My company is reaching for AI on much more than this, and of course it's just a personal anecdote, but throwing AI generated code + documentation into our codebase en masse has reduced our understanding of the system as a whole, it's caused a lot of "rubber stamping" on PR reviews, and it's introduced a layer of complexity that we're now going to deal with forever (provided we continue to use AI for everything). Your mileage may vary, my company doesn't adopt what I would consider "good engineering practices".

I'm hopeful that in a few years we can truly settle on the best use cases for AI, rather than considering it an everything tool. Not saying it cannot do a LOT, but from my experience I have found that it just cannot do everything (but we sure are trying).

AI is great at pattern matching many similar-but-oh-so-very-slightly-different cases.

Amazing for systematic refactoring that has a predictable outcome, but would need an ungodly amount of manual labor.