Think about a smart thermostat in your home. It doesn't just follow a fixed schedule. It senses the room's temperature, notices if someone is home, and even learns your daily routine. Based on all this information, it decides when to turn the heat or air conditioning on or off. This thermostat is a simple example of an AI agent.

An agent is anything that can be viewed as perceiving its environment through sensors and acting upon that environment through actuators. This is the core idea. It's not just a piece of code sitting idle; it's an active participant in some world, whether digital or physical.

Agents are the fundamental building blocks of AI systems. They are the doers, the entities that turn intelligence into action. Whether it's a robot navigating Mars, a spam filter clearing out your inbox, or a character in a video game, they all operate on this same basic principle.

Every agent follows a simple loop: it perceives, it thinks, and it acts. To do this, it needs a few key components.

Sensors are the agent's tools for perception. They gather information from the environment. For a self-driving car, sensors include cameras, GPS, radar, and microphones. For a spam filter, the sensors read the content of an incoming email, the sender's address, and other metadata.

Actuators are how the agent takes action. They are the mechanisms that affect the environment. The self-driving car's actuators are the steering wheel, accelerator, and brake pedal. The spam filter's main actuator is the command to move an email to the spam folder.

So, how does an agent decide what to do? The connection between what an agent perceives and what it does is called the agent function. It's the internal logic, the 'brain' that maps a sequence of perceptions to a specific action.

This function is the core of the agent's intelligence. For any given history of things it has sensed (a percept sequence), the agent function tells it which action to perform next.

In this notation, $f$ is the agent function. $P^*$ represents the entire sequence of percepts the agent has ever received, and $A$ represents the set of all possible actions the agent can take. The function essentially says, "Given this history of perceptions, here is the action you should take."

For our smart thermostat, a percept might be (22°C, 3 PM, no one home). The agent function would take this input and its memory of past percepts, and output an action like (lower temperature).

Understanding this basic structure of perception and action is the first step to seeing how even the most complex AI systems work.