Agents are systems built around large language models, not just smarter models themselves. They possess specialized components that enable them to persist and act beyond a single interaction. The fundamental structure of most agents includes a model for reasoning, instructions for guidance, memory for storing information, tools for expanding capabilities, and an execution loop for continuous operation. The model, or LLM, acts as the reasoning engine, processing language and deciding on actions. Instructions define the agent's purpose, identity, and desired behavior, focusing its decision-making. Unlike an LLM's context window, which resets with each turn, an agent's memory is an external, persistent store of information accessible across sessions. This memory allows for operational continuity, maintaining context over time. Tools provide agents with functionalities beyond text generation, enabling them to interact with the external world. The execution loop orchestrates the agent's cycle of observation, decision-making, action, and evaluation. Without any of these core components, an agent's behavior is significantly altered. The distinction between an LLM's apparent memory within a conversation and an agent's persistent memory is crucial for understanding agent functionality. The system, not the model itself, manages and feeds information into the LLM's context window. Ultimately, an agent's effectiveness relies on the synergistic interplay of all its constituent parts.