Reflections on Three Years of Building LangChain

Key takeaways

1. LangChain began as a side project built from patterns Chase saw people using at meetups, and he says he had no idea how big LLMs would become.
2. Chase says LangSmith was built as a separate product, neutral to any LLM or framework, so it could be best-in-class even for developers who never used langchain.
3. Negative feedback in summer 2023 included fixable issues like breaking changes, hidden prompts and outdated docs, plus a harder one: people wanted more control. Chase says easy high-level interfaces got in the way of customizing for production.
4. Chase says that criticism led to LangGraph, built around controllability (no hidden prompts or context engineering) and a first-party production runtime with streaming, statefulness, human-in-the-loop and durable execution.
5. When rebuilding langchain for 1.0, the team accepted large breaking changes but kept 0.x alive as langchain-classic with extended support.