I agree with most of the comments here, langchain is good when you’re getting started. Doc isn’t great but has enough breadcrumbs if you’re willing to get into the source code. Prompts are key and you can pull from them as “starter templates”. The model abstraction was appealing to me at the beginning but you need to adjust prompts based on different models so that becomes a little tedious.
The vector database and memory abstraction is useful but takes some langchain specific memory management understanding to make sure the memory follows you.
For example I was using local memory then easily swapped it out for an open source solution, zep (getzep.com) for persistent memory. Also swapping vector databases and other tools is pretty easy.
If you’re new to AI development trolling the document and seeing what tools are out there was really helpful.