LangChain

We're selective about LangChain. For complex agent orchestration with a lot of chained tool calls, retrievers, and memory, it can save real time. For most production AI features — a RAG pipeline, a classification endpoint, a structured generation task — it's heavier than necessary and introduces an abstraction layer we don't need.

When we use LangChain, we use it deliberately. When we don't, we write the orchestration directly and usually find it clearer to read and debug. We'll work in LangChain if a client's codebase is already built around it, but we don't pitch it as our default.