To take advantage of AI opportunities, people often upload entire documents into AI systems, and AI analyzes the material as one solid piece of information. That’s useful to some degree. But the approach is like asking AI to process a wall of data: There is no context, no appreciation of how individual pieces of knowledge within the document fit together. We create opaque data containers, so to speak, that make the AI inefficient.
To break through such barriers, AI needs to see information in the document as individual, internal knowledge units — as discrete paragraphs, tables and the like — rather than see only the document’s external structure. A standard approach to Retrieval Augmented Generation, or RAG, draws from trusted external sources that provide even more context and reduce AI hallucinations and other inaccuracies. A variation of that — agentic RAG — goes further.
“Think of standard RAG as being a librarian,” said Mike Gilger, Chief Technology Officer at Modus Operandi. “Yeah, it can grab a book and give it to you, but agentic RAG actually adds a whole staff to it, a research staff, so it validates that it [gives you] something that’s reasonable…. It brings the RAG technology to something that’s much more usable in a business environment.”
In this video interview, Gilger discusses how agencies can improve their AI outcomes with more reliable, contextual data. Topics include:
- Why agencies struggle to derive useful, verifiable AI results
- How agentic RAG works and the specific benefits it offers
- How AI helps employees leave workflow drudgery behind