langgraph’s killer for llm-savvy teams needing complex, stateful workflows, think multi-agent systems or cyclic graphs. it’s faster than custom pipelines, with langsmith debugging and one-click scaling. but it’s steep to learn, and debugging multi-agent setups can suck if state’s messy. langchain’s better for simple, linear tasks; langgraph’s for intricate control. autogen’s easier but less precise; crewai’s rigid. langgraph’s your pick for dynamic, production-ready projects.

Not really.

Langchain was designed for chaining requests together, back when you had to maintain context with AI yourself. OpenAI figured a fix that works for both humans and AI - putting comms in a chat format, and so we didn't need to chain input.

Langchain was one of the earliest to RAG, but it was more of a hack and didn't do as well as many alternatives.

It seems like they're going into agents now. No comment but it feels like they're a step behind.