Agentic Artificial Intelligence (Agentic AI) represents a paradigm shift from traditional rule-based and task-specific AI systems toward autonomous, goal-driven intelligent systems capable of reasoning, planning, and acting in dynamic environments. Unlike conventional AI agents that operate within predefined constraints, Agentic AI systems leverage foundation models, particularly large language models (LLMs), to achieve adaptive behavior, persistent memory utilization, and multi-agent coordination. This paper provides a comprehensive review of Agentic AI, covering its evolution from symbolic AI to modern generative and agent-based systems. Core architectural components, including perception, memory, planning, execution, and orchestration layers, are analyzed alongside prominent frameworks such as AutoGen, CrewAI, MetaGPT, LangGraph, and Semantic Kernel. A structured taxonomy of agentic systems is presented based on architecture, interaction paradigms, and domain specialization. The paper further examines enabling technologies such as reinforcement learning, multi-agent systems, and neurosymbolic AI, and discusses methodologies including BDI-based reasoning frameworks. Applications across healthcare, finance, manufacturing, and human–AI collaboration are explored to highlight practical impact. Finally, critical challenges including scalability, evaluation limitations, computational complexity, and ethical concerns are analyzed, followed by future research directions focusing on explainability, adaptive learning, and cross-domain generalization. This review aims to provide a coherent and analytical understanding of Agentic AI and its potential to redefine intelligent systems.
Sharma et al. (Wed,) studied this question.