Agentic Retrieval Augmented Generation: A Compact Review

Abstract

Retrieval-Augmented Generation (RAG) enhances large language models (LLMs) by incorporating external knowledge during inference, improving factual accuracy and contextual relevance. However, conventional RAG systems often struggle with complex, dynamic, or multi-step tasks due to their static retrieval- generation pipelines. Agentic RAG addresses these limitations by embedding autonomous agents into the RAG framework, enabling capabilities such as planning, tool use, memory integration, and adaptive decision-making. This review outlines the architecture of Agentic RAG, highlights its advantages in dynamic applications like long-form question answering and research automation, and discusses challenges such as latency, control, and ethical design.