quoting what I said in that thread 🙃 having used LangChain, I found it to be far more harm than help. the abstractions are clunky and get in the way. I saw it put on twitter somewhere like "LangChain is predicated on the idea that LLMs are hard, and they're working to make it harder" -- that resonated with my experience
I would rather just work with strings directly (at the end of the day it's mostly just formatting strings for you) or use a more Pythonic framework