ScholarWorks@성균관대학교

초록

Considering that the field of Artificial Intelligence (AI) is shifting toward autonomous, goal-oriented systems, there is a need for a systematic overview of the emerging Agentic AI landscape. We attempt to provide a survey of Agentic AI presented and examined by large-scale models, demonstrating the foundational architectures, diverse applications, and inherent technical challenges of these systems. We propose a robust multidimensional taxonomy that classifies agents based on their structural design, autonomy levels, application domains, and sustainability with resource efficiency. We also provide an understanding of the operational principles of several recent open-source frameworks and a comparative analysis of design patterns that facilitate scalable and high-performance deployment. We then examine several future directions for agentic AI systems, including robustness, safety, resource efficiency, and long-horizon planning.

키워드