Source: https://arxiv.org/abs/2510.09244

Overview of the paradigm shift from traditional Large Language Models (LLMs) to Agentic LLMs, defining the latter as autonomous, goal-oriented systems designed to overcome the limitations of passive, stateless LLMs.

It details the agentic architecture, which is based on four integrated components—Perception, Reasoning, Memory, and Execution—that allow the AI to interact with and act upon the external world.

The text contrasts the reactive nature of traditional LLMs with the proactive, problem-solving capabilities of agents, exploring practical applications across sectors like healthcare, finance, and robotics.

Finally, the report addresses the significant technical and ethical challenges, such as state desynchronization and accountability, and outlines future trends, including the move toward multi-agent systems and smaller, specialized models.