Reasoning RAG via System 1 or System 2: A Survey on Reasoning Agentic Retrieval-Augmented Generation for Industry Challenges

TL;DR

The paper addresses the limitations of static RAG pipelines in industrial, knowledge-intensive tasks. It surveys Reasoning Agentic RAG, partitioned into predefined reasoning (fixed modular pipelines) and agentic reasoning (autonomous tool-use driven by reasoning). It catalogs representative methods across both curricula, including RAGate, Self-RAG, RAPTOR, Adaptive-RAG, and Modular-RAG in the predefined category, and ReAct, Self-Ask, function calling, Search-R1, DeepRetrieval, and DeepResearcher in the agentic category, with discussions of architectures, reasoning strategies, and training signals. It identifies key challenges and future directions toward flexible, robust, and industry-ready systems, and provides a GitHub collection of related work.

Abstract

Retrieval-Augmented Generation (RAG) has emerged as a powerful framework to overcome the knowledge limitations of Large Language Models (LLMs) by integrating external retrieval with language generation. While early RAG systems based on static pipelines have shown effectiveness in well-structured tasks, they struggle in real-world scenarios requiring complex reasoning, dynamic retrieval, and multi-modal integration. To address these challenges, the field has shifted toward Reasoning Agentic RAG, a paradigm that embeds decision-making and adaptive tool use directly into the retrieval process. In this paper, we present a comprehensive review of Reasoning Agentic RAG methods, categorizing them into two primary systems: predefined reasoning, which follows fixed modular pipelines to boost reasoning, and agentic reasoning, where the model autonomously orchestrates tool interaction during inference. We analyze representative techniques under both paradigms, covering architectural design, reasoning strategies, and tool coordination. Finally, we discuss key research challenges and propose future directions to advance the flexibility, robustness, and applicability of reasoning agentic RAG systems. Our collection of the relevant research has been organized into a https://github.com/ByebyeMonica/Reasoning-Agentic-RAG.

Figures (6)

• Figure 1: Overview of two major types of reasoning Agentic Systems.
• Figure 2: Distributed Works of Reasoning Agentic RAG.
• Figure 3: A taxonomy of Reasoning Agentic RAG.
• Figure 4: Illustration of the evolution of Reasoning Agentic RAG.
• Figure 5: A demonstration of Predefined Reasoning.