<p>Agentic Artificial Intelligence (AI) marks a shift from traditional AI systems that simply generate responses to autonomous systems that can independently plan to achieve goals with minimal human intervention. These models can do much more than just respond to prompts as they can observe, adapt, coordinate with other agents and even refine their own outputs over time. The present study draws insights from fifty-one recent studies to understand how agentic AI is being built and used now a days. Agentic AI systems appear in the domains of healthcare, digital twin architectures, educational platforms, e-commerce applications, cybersecurity systems and large-scale network management systems. They often improve efficiency, reduce manual workload, and help in making more informed decisions. However, this increased autonomy also raises several concerns. This is because autonomous systems that can act without human intervention must be reliable, explainable, secure and aligned with human expectations. Many implementations of such systems are still in early stages, lacking standard evaluation methods and are facing challenges such as data access, ethical responsibility, and coordination among multiple agents. For clearer understanding, this study outlines a taxonomy of agentic AI and describes its current application domains, discusses common architectures and techniques, and highlights its limitations and future directions. The results of this study suggest that progress in governance, multi-modal reasoning and scalable coordination will be central to advancing safe and useful agentic AI systems.</p>