Spent the last year building LLM features in production. Most "the AI is wrong" bugs turned out not to be model problems at all. Wrote up 10 lessons - prompt engineering, evaluation, RAG, agentic workflows, and where MCP fits in.
Full article here: https://dev.to/sureshbabunarayanan/10-hard-earned-lessons-from-a-year-of-building-with-llms-in-production-akc
