Which is better for building AI agents in 2024: LangChain or CrewAI?

All answers to this question.

I’ve used both extensively. CrewAI is fantastic for "collaborative" tasks—think of it as a manager delegating to a team. It’s built on top of LangChain, so it’s not necessarily an "either-or" situation. CrewAI simplifies the "process" (sequential or hierarchical) but it can be less flexible if you need very granular control over the prompt or the tool-calling loop. If you are building a complex, custom enterprise app, stay with LangChain/LangGraph. If you want to spin up a "content marketing team" agent in 20 minutes, CrewAI is significantly faster to implement.

   Answered 2024-07-17 by Sharon Perry

Is it possible to wrap a CrewAI "crew" inside a LangChain chain if I want to use LangSmith for debugging and observability?

   Answered 2024-07-20 by Edward King

• Edward, yes you can! Since CrewAI uses LangChain under the hood, you can just set your LANGCHAIN_TRACING_V2 environment variables to 'true'. Every task and delegation within your "Crew" will show up in LangSmith. It’s actually the best way to debug where your agents are losing the plot during a multi-stage workflow. I do this for all my production RAG pipelines now.

     Commented 2024-07-21 by Richard Cooper

CrewAI’s "manager" feature is great, but I found it sometimes gets stuck in infinite loops. LangChain's LCEL gives me more confidence for high-traffic applications.

   Answered 2024-07-22 by Joseph Parker

• Valid point, Joseph. LCEL (LangChain Expression Language) is definitely more verbose, but that explicitness makes it much easier to unit test individual components of your AI agent.

     Commented 2024-07-23 by Rebecca Hayes