AI Agents in Research Automation: A Comprehensive Review

This review explored the emerging idea of agentic research engines as a step forward from traditional RetrievalAugmented Generation (RAG) systems. RAG architectures have certainly improved how information can be accessed and summarized by combining retrieval with language generation. That improvement is real. Yet the approach still carries some clear limitations. Many RAG systems rely on static knowledge bases and relatively shallow reasoning steps, and they often struggle to adapt when the information landscape shifts quickly. Because of this, their usefulness starts to fade when the task moves beyond simple question answering and into more demanding research work. Tasks that involve iterative exploration, comparison across multiple sources, or careful verification of conclusions often expose these weaknesses. Looking through recent literature, one can see why interest in agentic AI architectures has grown. These systems introduce autonomous workflows built around multiple specialized agents that cooperate during the research process. Instead of simply retrieving documents and summarizing them, such architectures include components responsible for planning tasks, gathering sources, reasoning through evidence, and verifying conclusions. Shared knowledge stores and feedback loops allow the system to revisit earlier steps and refine its outputs. In effect, the design begins to resemble the way a human researcher might approach a problem: gather evidence, reconsider assumptions, and occasionally revise conclusions. That shift from passive retrieval toward active reasoning opens the possibility of deeper analysis, fresher information coverage, and clearer reasoning traces. Still, the approach is not without its difficulties. Questions around evaluation methods, computational cost, scalability, and governance remain unresolved and deserve careful attention [2], [3].

[1] P. Lewis et al., “Retrieval-augmented generation for knowledge-intensive nlp tasks,” Advances in Neural Information Processing Systems, 2020. [2] J. Schneider, “Generative to agentic ai: Survey, conceptualization, and challenges,” arXiv preprint arXiv:2504.18875, 2025. [3] D. B. Acharya, K. Kuppan, and B. Divya, “Agentic ai: Autonomous intelligence for complex goals—a comprehensive survey,” IEEE Access, 2025. [4] Q. Wu et al., “Autogen: Enabling next-gen llm applications via multiagent conversation,” arXiv preprint arXiv:2308.08155, 2023. [5] S. Hong et al., “Metagpt: Meta programming for multi-agent collaborative framework,” arXiv preprint arXiv:2308.00352, 2023. [6] A. Madaan et al., “Self-refine: Iterative refinement with large language models,” arXiv preprint arXiv:2303.17651, 2023. [7] S. Russell and P. Norvig, Artificial Intelligence: A Modern Approach, 4th ed. Pearson, 2021. [8] V. Ingale, B. Wankar, K. Jadhav, T. Adedoja, V. K. Borate, and Y. K. Mali, “Healthcare is being revolutionized by ai-powered solutions and technological integration for easily accessible and efficient medical care,” in 2024 15th International Conference on Computing, Communication and Networking Technologies (ICCCNT). IEEE, 2024, pp. 1– 6. [9] A. K. Pati, “Agentic ai: A comprehensive survey of technologies, applications, and societal implications,” IEEE Access, 2025. [10] J. Wei et al., “Chain-of-thought prompting elicits reasoning in large language models,” Advances in Neural Information Processing Systems, 2022.