Agentic Retrieval-Augmented Generation: A Survey on Agentic RAG
by Singh, Ehtesham, Kumar, Khoei & Vasilakos (arXiv)
The reference survey mapping how autonomous agents make RAG dynamic, reflective, and multi-step.
Overview
"Agentic Retrieval-Augmented Generation: A Survey on Agentic RAG" (arXiv:2501.09136) by Aditi Singh, Abul Ehtesham, Saket Kumar, Tala Talaei Khoei, and Athanasios V. Vasilakos surveys how classic RAG is evolving into Agentic RAG, where autonomous agents embed agentic design patterns—reflection, planning, tool use, and multi-agent collaboration—into the retrieval loop to manage retrieval dynamically and refine understanding iteratively. First submitted January 15, 2025 and revised through April 2026, the paper introduces a principled taxonomy that organizes Agentic RAG architectures along four dimensions: agent cardinality (single vs. multiple agents), control structure (from sequential to adaptive collaboration), autonomy level, and knowledge representation. It compares design trade-offs across existing frameworks, reviews applications in healthcare, finance, education, and enterprise document processing, and identifies open research gaps in evaluation methodology, agent coordination, memory systems, computational efficiency, and governance. It is a strong grounding reference for engineers designing retrieval systems that go beyond static, single-shot retrieval.
At a Glance
- Topic
- RAG
- Level
- Advanced
- Format
- Paper
- Cost
- Free
- Duration
- ~1-2 hour read (survey)
- Provider
- Singh, Ehtesham, Kumar, Khoei & Vasilakos (arXiv)
- Hands-on
- No
- Certificate
- None
What You’ll Learn
- ✓How Agentic RAG differs from and improves on static single-shot RAG
- ✓A taxonomy of agentic RAG architectures by agent cardinality, control, autonomy, and knowledge representation
- ✓Where reflection, planning, tool use, and multi-agent collaboration fit in a retrieval loop
- ✓Design trade-offs across current agentic RAG frameworks
- ✓Open problems in evaluation, coordination, memory, efficiency, and governance
Highlights
- •Widely-cited reference survey on Agentic RAG
- •Clear four-dimension taxonomy for classifying architectures
- •Covers real applications across healthcare, finance, education, and enterprise
- •Kept current through 2026 revisions
Who It’s For
Best For
- ✓AI engineers designing retrieval systems beyond basic RAG
- ✓Researchers surveying the agentic RAG landscape
- ✓Architects evaluating single- vs. multi-agent retrieval designs
Prerequisites
- •Working knowledge of RAG and retrieval pipelines
- •Familiarity with LLM agents
FAQ
What is Agentic Retrieval-Augmented Generation: A Survey on Agentic RAG?
A comprehensive academic survey, for engineers and researchers, of Agentic RAG—RAG pipelines augmented with autonomous agents that reflect, plan, use tools, and collaborate. It provides a structured taxonomy of architectures and design trade-offs.
Is Agentic Retrieval-Augmented Generation: A Survey on Agentic RAG free?
Agentic Retrieval-Augmented Generation: A Survey on Agentic RAG is free to access.
What level is Agentic Retrieval-Augmented Generation: A Survey on Agentic RAG for?
Agentic Retrieval-Augmented Generation: A Survey on Agentic RAG is aimed at a advanced audience. Recommended background: Working knowledge of RAG and retrieval pipelines, Familiarity with LLM agents.
How long does Agentic Retrieval-Augmented Generation: A Survey on Agentic RAG take?
Expect roughly ~1-2 hour read (survey). Most learners work through it at their own pace.
What will I learn from Agentic Retrieval-Augmented Generation: A Survey on Agentic RAG?
You'll learn: How Agentic RAG differs from and improves on static single-shot RAG; A taxonomy of agentic RAG architectures by agent cardinality, control, autonomy, and knowledge representation; Where reflection, planning, tool use, and multi-agent collaboration fit in a retrieval loop; Design trade-offs across current agentic RAG frameworks; Open problems in evaluation, coordination, memory, efficiency, and governance.