Agentic Artificial Intelligence (Agentic AI): Fundamentals, Architectures and Applications

Agentic Artificial Intelligence (Agentic AI) represents a significant evolution from traditional reactive and predictive AI systems by introducing computational systems capable of acting autonomously, being goal-oriented, reasoning and planning activities, and dynamically adapting to new scenarios. This article presents an analysis of the conceptual foundations of Agentic AI, with a brief introduction and the main characteristics of agentic systems, such as goal orientation, autonomy, reasoning and planning, proactivity, and adaptation. Next, the most common architectures of agentic systems are examined. The article also explores four of the main frameworks widely used today in the development of agentic systems, such as LangChain, AutoGen, CrewAI, and LangGraph, highlighting their approaches and limitations. Finally, the ethical challenges associated with autonomous decision-making are addressed, with emphasis on the need for transparency, explainability, and responsible governance, especially in critical contexts where agentic systems can affect human life or the environment.