Agentic RAG — Retrieval as a Tool, Not a Reflex

1. What is the main criticism of the fixed pipeline stage in the naive 'retrieve-then-generate' approach?

It always provides accurate answers. It biases the model with irrelevant context. It leads to faster processing times. It eliminates the need for multiple searches.

2. What does the agent's routing policy include for handling precise questions?

Only semantic index retrieval. Direct reads and reference-following. Only lexical searches. Only multi-hop reasoning.

3. Why is agentic RAG favored over a simple retrieval model?

It relies solely on predefined indexes. It can assemble answers spanning multiple sources. It simplifies the model architecture drastically. It requires no user input for processing.

4. What should be prioritized over defaulting to the semantic index according to the lesson?

Fetching random snippets. Searching files for specific symbols. Performing fuzzy searches. Using a precomputed index exclusively.

Continue: Why Multiple Agents → Review this lesson Retake quiz