I do some Rails work and I’ve found that Ruby and Rails are one of generative AIs weaker languages. I usually get pretty shoddy output[1].
I generally don’t use AI during Rails stuff, other than as a hint for a Google search or a docs lookup.
Are you having a different experience?
1. Shoddy output with Ruby and Rails knowledge. In general, I would consider myself pretty advanced with “prompt engineering”.
I'm not expecting perfect output, if there is such a thing. I'm expecting output equivalent to a junior engineer or intern at this point in time. I can tell an agent in the editor I want it to add a many to many join relationship and a multi select field to a form and it does it pretty decently. In some ways this is more automating the tedious tasks which is really what AI's strong suit is right now, in my opinion. Sure I could go move a method around, refactor the 5 classes that call it, etc. but I can delegate that task and its done in a few seconds while I think about the next thing I want to do.
I find if you get too lengthy with tasks or too advanced AI goes off the rails so to speak, sorry for the pun.
I've found that OpenAI models write fantastic Rails code, having a deep understanding of Rails codebase conventions. Anthropic's models do seem to have a very hard time writing Ruby in general.
Similar experience. Although, recently just tried Claude Code and it seems to be a pretty good upgrade (both Sonnet and Opus). I'd suggest you to give it a try if you haven't. For UI, Playwright MCP helps a lot. And since it can run rspec too, it can get faster feedback.
To me it is better now, but not as good as certain languages. Since that I'm using Go as well, I do notice Claude Code perform better with Go.