Hi guys, do you know if the prefect developers gav...
# ask-community
g
Hi guys, do you know if the prefect developers gave up on ControlFlow? I see the last release was on Feb 6 and they seem to be pushing FastMCP instead. I really liked the approach with ControlFlow which just builds on the serverless / stateless paradigm of Prefect, what happened?
n
tldr; all the useful stuff controlflow had now lives in marvin, which we still passively maintain marvin is also built on pydatnic-ai instead of langchain
g
Thanks for the info! Can you also give me a TLDR on why you guys prefered to build on top of pydantic-ai instead of langchain?
n
• same broad LLM support, more lightweight, better typing, overall better DX (in our opinion) • we have lots of experience working directly with the pydantic team on things (in prefect, previous marvin versions etc)
g
That makes sense! Thanks for the reply. I have been checking this out and looks super cool tbh. I was thinking for my team's use case to add langchain + lcel to our prefect flows that will be agentic, but now this makes me re-think. Marvin might do exactly just that (which is why I was already exploring ControlFlow to begin with).
šŸ‘ 1
n
just to clarify, id say fastmcp and marvin are not an "either/or" pair of tools marvin is for structured outputs and agentic while loops with broad LLM support fastmcp is for writing MCP servers and clients, which very well might use structured outputs or agentic while loops on the inside
šŸ‘ 1
g
right, that makes total sense
catjam 1
Can I ask one final question? From my recent understanding, Pydantic AI is basically a full blown replacement for langchain + langgraph already, right? I don't have a huge amount of experience with this just yet. So this may sound stupid or naive, but what exactly is Marvin adding on top of that? Maybe you can give me a use case where Marvin would shine compared to just using Pydantic AI. Sorry if it's a stupid question and thanks in advance for your time šŸ™
g
thanks so much!