Towards Trustworthy and Cost-Efficient Data Integration: From Naïve RAG to Agentic RAG

TL;DR - This paper surveys the evolution from conventional RAG to autonomous, multi-agent RAG for enterprise data integration. It outlines how knowledge grounding and adaptive retrieval could improve verifiability, scalability, and cost efficiency.

• Compares classic RAG with GraphRAG and knowledge graph-based RAG for bridging parametric and contextual knowledge.
• Frames Agentic RAG as agents that dynamically plan, retrieve, refine, and reason across complex integration tasks.
• Defines trustworthiness through evidence-grounded, transparent decisions that resist hallucinations and remain consistent.
• Discusses computational bottlenecks, optimization strategies, and open challenges rather than reporting new empirical results.