Retrieval-augmented generation (RAG) is a technique where an AI model fetches relevant documents from an external knowledge source at query time and uses them to generate a grounded answer. Instead of relying solely on what the model learned during training, RAG retrieves fresh, specific context and conditions its response on it.
A typical RAG pipeline turns a query into a vector, searches a knowledge base for the closest matching passages, and passes those passages to the model as context. The benefit is current, citable, source-anchored answers. The limit is that RAG is only as good as the knowledge base behind it.
That limit is worth naming plainly. A RAG agent bound to a knowledge base hits a ceiling on the questions a KB never documented, the long-tail intents that make up much of real support volume. Retrieving a passage is not the same as knowing whether the AI should act on this specific intent at all.
Aide, the agentic AI platform for customer experience, uses retrieval where it fits, but leads with intent-first architecture: a custom intent classifier and a three-level Customer Intent Map decide what gets automated, intent by intent, before any answer is generated. Each intent is test-gated on real historical conversations in the Agent Simulator, so retrieval errors are caught before deploy, not after. Grounded by intent, not bound by a knowledge base.
Frequently asked questions
- What is the difference between RAG and fine-tuning?
- RAG fetches external knowledge at query time and leaves the model unchanged, which makes it easy to update and audit. Fine-tuning bakes knowledge into the model's weights through additional training. RAG is generally preferred when facts change often or must be citable.
- Why is RAG not enough for customer support automation?
- RAG retrieves answers but does not decide which customer intents should be automated, or verify that automation is safe before it ships. Intent-first architecture adds that governance layer: classify the intent, test it, then deploy only what passes.