The real value of agents is not just basic search/RAG (we can agree with that), but to automate domain-specific knowledge work and processes. I made a simple example of this for auto-insurance claim processing, in the latest edition of my series on agentic document workflows. Given an insurance claim, parse out relevant information. Find relevant policy guidelines as well as policyholder-specific declarations through retrieval as just one step in this overall process. Use it to generate a structured policy recommendation and claim decision on how much to settle for. Check it out: github.com/run-llama/llamacl… It uses LlamaCloud which we’re opening up soon beyond a waitlist, but in the meantime you can use any vector index for your docs. LlamaCloud signup: cloud.llamaindex.ai/