6 posts found.
The should-EMs-code debate is a false binary. The builder-manager model — coding to understand, not to ship — is the third path that survives AI flattening, earns team trust, and creates organizational leverage.
AI coding agents are great at generating new code but catastrophically bad at long-term maintenance. I tested them on real codebases — here is what actually breaks, why, and the engineering guardrails that make them usable.
MCP is the protocol connecting AI agents to the real world, but most tutorials stop at hello world. Here is how to build MCP servers that survive production — with auth, error handling, observability, and the patterns I wish existed when I started.
I needed my AI assistant to answer architecture questions, but every coding tool I tried understood individual files and failed at the big picture. So I built a system that auto-generates comprehensive architecture documents from code repositories using hybrid AST extraction and LLM synthesis. Here is how it works and what I learned.
I shipped a naive RAG pipeline and called it done. Here is what actually broke in production, and the multi-stage architecture that took retrieval precision from 58% to 91%.
AI coding assistants are powerful, but using them without structure leads to unmaintainable code, eroded skills, and team dysfunction. Here is how to use AI deliberately.