Agentic Workflow for Automated Systematic Literature Review: A Case Study on Generative AI Applications in the Oil and Gas Industry

Abstract

The rapid expansion of generative AI across the oil and gas applications has created an urgent need for structured review on emerging research trends. However, traditional systematic literature reviews (SLRs) are time-consuming and difficult to scale as the volume of AI publications grows every day. This paper proposes a novel agentic, multi-agent workflow that automates the SLR process and evaluates the maturity of generative AI applications in the upstream sector, while enabling the review to be continuously updated with newer publications. The methodology begins with defining domain-specific research questions and performing a targeted keyword search. A set of inclusion/exclusion criteria is implemented within a Python-built agentic pipeline. The system uses ReAct-based reasoning agents, content-routing for task specialization, and an orchestrator-worker architecture that coordinates paper screening, abstract interpretation, and classification. Validation gates and an evaluator-optimizer loop ensure consistency, minimize hallucination, and maintain reproducibility. The workflow follows a hybrid human-in-the-loop design, where screening and interpretation are automated by the agentic system, while technical rigor and domain relevance are validated and reviewed manually. The workflow processed 200+ initial records, automatically shortlisted papers relevant to the upstream industry, and categorized them into methodological clusters including Transformers, Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), AI Agents, Variational Autoencoders, Diffusion Models, and Hybrid Architectures. Comparative analysis revealed that generative AI is most mature in knowledge retrieval and operations advisory, while applications in reservoir management and well-construction planning remain in early development. The automated pipeline reduced manual efforts, demonstrating significant efficiency gains over traditional review methods. Beyond improving efficiency, the framework establishes a scalable and continuously updatable foundation for monitoring the rapid evolution of generative AI research in the oil and gas industry. The results also highlighted the need for careful data governance, model traceability, and responsible deployment practices to ensure reliable adoption. This paper implemented an agentic AI workflow for conducting a systematic literature review of generative AI applications in the oil and gas industry. The results demonstrated that multi-agent systems can significantly accelerate the review process and provide a scalable, continuously updatable foundation for future research assessments in digital oilfield innovation.