Textbook of AI

An Agent is an entity that perceives its environment through sensors and acts upon it through actuators to achieve goals. This abstraction, central to the Russell–Norvig framework for AI, encompasses everything from a thermostat (a reactive agent with a single sensor and actuator) to a self-driving car (a complex agent with cameras, lidar, GPS, and motor controls) to a conversational AI assistant (an agent that perceives text input and acts by producing text output or calling tools).

Agents are typically classified by their architecture and capabilities. Simple reflex agents act purely on the current percept via condition–action rules. Model-based reflex agents maintain an internal model of the environment. Goal-based agents have explicit goals and plan to achieve them. Utility-based agents optimise a continuous measure of preference. Learning agents improve their behaviour through experience. Rationality—choosing actions that maximise expected utility given available information—is the gold standard for agent design.

In the 2020s, the term "AI agent" has come to refer specifically to systems built on large language models that can plan, call tools, execute code, browse the web, and maintain memory across interactions. Frameworks such as ReAct (Reasoning and Acting) interleave thought and action to tackle complex multi-step tasks. Agentic AI represents a qualitative shift from models that merely generate text to systems that take consequential actions in the world, raising fresh questions about safety, control, and accountability.