Hey everyone,
I've been using Cursor as my daily driver for a few months now, and I really appreciate its core workflow for mainstream languages. The agent mode and the chat-driven refactoring feel like magic when I'm working in TypeScript or Python. However, I've hit a consistent and pretty significant roadblock whenever I try to bring it into my Elixir projects.
The issues feel systemic, not just occasional hiccups. It seems to really struggle with the functional programming paradigm and the specific idioms of the BEAM ecosystem. For instance, when I ask it to help me refactor a Phoenix context module, it often suggests patterns that would be more at home in an object-oriented language, missing the point of `Ecto` changesets and pipe operators entirely. Its suggestions for `GenServer` callbacks are frequently syntactically incorrect or ignore OTP best practices.
Even more frustrating is its handling of Elixir's macro system and DSLs, like those in Ash Framework or Surface UI. It either fails to recognize the syntax entirely or generates code that's completely invalid. The "chat with your codebase" feature also seems less effective, as it appears to have a weaker foundational understanding of Elixir's standard library and common library structures compared to, say, JavaScript's npm ecosystem.
I'm curious if others have run into similar walls with other less-mainstream tech stacks—maybe Rust's unique ownership model, or niche frontend frameworks like SvelteKit? Is this a training data problem, where the models powering Cursor simply have less exposure to these ecosystems? Or is there a configuration or setup step I'm missing to better orient it toward a functional language?
I want to love it for all my work, but right now, it feels like I have to switch back to a simpler editor for my Elixir work, which defeats the purpose of having an "AI-first" editor. Any shared experiences or insights would be super helpful.
— Alex
Let's keep it real.