Abstract Agentic Artificial Intelligence (Agentic AI), as a new generation of intelligent systems, extends beyond mere text or image generation by incorporating components such as multi step reasoning, persistent memory, multi agent interaction, and purposeful tool use. These features enable autonomy and dynamic decision making in open and complex environments. This paper provides an extensive review of the existing literature on Agentic AI. To this end, the conceptual and historical distinctions between Agentic Artificial Intelligence, classical agents, and general-purpose language models are first examined. Subsequently, the proposed architectural framework encompassing perception, role adoption, memory, planning and reflection, action and tool use, and online learning is elaborated. The discussion then turns to key infrastructures, including state and memory management, orchestration and Machine Learning Operations (MLOps), as well as agent safety. Evaluation metrics and protocols are reviewed, ranging from interactive and long horizon testing to process level assessments. Furthermore, applications of Agentic AI are introduced in areas such as scientific discovery, intelligent education, finance, and robotics, followed by an analysis of core challenges including emergent behaviors, multi agent security, accountability, and energy sustainability. The findings of this review suggest that integrating methods such as Agentic Retrieval Augmented Generation (Agentic RAG), agent chaining, and formal threat modeling can provide the foundation for the next generation of trustworthy and scalable systems. Finally, research gaps and future directions are identified, including the design of reproducible evaluation protocols, multi agent safety frameworks, and the integration of online learning with long term memory, which are presented as primary priorities for research and development in this field.
Peykani et al. (Mon,) studied this question.