I’m early in my journey, both with actually attempting to write robust Elixir/OTP, and with LLM’s, but here are a few tips that have helped already:
- Hook up context7 MCP in Cursor. Then, in every new agent chat, instruct the agent to use context7 to read the latest Elixir docs.
- Iterate on your docs. Focus on standards. Use dialyzer heavily. Keep warnings in check. Develop a robust strategy for OTP. Use the LLM’s themselves to develop these standards for your document set, leaning on your knowledge of BEAM/OTP principles.
- Be willing to throw away an attempted implementation, using it for lessons learned in feedback loops with Claude/Gemini/Grok/etc.
I’m also focused on using this experience to enhance and accelerate learning Elixir/OTP development, rather than letting the LLM think without me.
LLM’s for coding are not going anywhere. It’s probably wise to start getting acclimated to using the latest AI tools, if you haven’t started already.
Bullish long term on Elixir for AI, both for development and as a platform. It’s still early.
