Requirements Engineering with Generative AI: A Systematic Literature Review of Using AI Agents for Requirements Engineering Automation

Requirements development is a critical phase of software development, as the requirements developed during this phase provide the foundation for subsequent project activities. Because software requirements are predominantly expressed in natural language, automation of requirements engineering (RE) activities has historically been challenging. Recent advances in the natural language processing capabilities of Large Language Models (LLMs), however, have created new opportunities for automating RE tasks through AI agents. Accordingly, research on LLM-supported, agent-based RE automation has grown substantially in recent years. This study presents the results of a literature review of research on LLM-supported agent-based automation of software requirements engineering. The review identified 50 studies, of which 24 propose or describe agent-based architectures for RE automation, while the remaining 26 introduce benchmarks, baselines, evaluation environments, or comparative approaches for assessing agent-based solutions. The 24 studies describing an agent-based architecture are classified according to the requirements activities defined in the Software Requirements knowledge area of SWEBOK V4.0a, namely requirements elicitation, analysis, specification, validation, and tracing. Our analysis identifies the dominant agent architectures employed across RE activities, recurring design patterns, evaluation approaches, and key limitations of agent-based RE systems. Although significant advances have been made, several critical challenges remain, including hallucinations, scalability, and the lack of standardized evaluation frameworks. These findings highlight important directions for future research toward reliable, scalable, and systematically evaluated AI agent-based automation of software requirements engineering.