Everyone's raving about building RAG POCs in a weekend. So I tried both. The hype is... thick.
LlamaIndex's "data framework" promise felt like more boilerplate than necessary. LangChain's "1000 tools" approach is a maze of abstractions. For a simple Salesforce data connector and a query interface, I spent more time wrestling with their respective 'agent' or 'query engine' concepts than writing actual logic. Simpler? Define simple. If you want "out-of-the-box" magic, prepare for hidden complexity when you inevitably need to customize. If you want control, you'll end up writing your own glue code anyway.
So, which one actually gets out of your way faster for a barebones, proof-of-concept pipeline? Or is the simpler move to skip both and use the LLM API directly with a basic script?
If it sounds too good, read the release notes