At its heart, an Artificial Intelligence system is all about making smart decisions. The entity that does the deciding and the acting is called an agent. Think of yourself. You sense the world around you (you see a red light, feel the cold), you process that information, and you take action (you stop the car, you put on a jacket). An AI agent does the same thing, just digitally.

The agent is the core of the system, but it doesn't exist in a vacuum. It operates within an environment, which is just a fancy word for the world or system it interacts with. For a self-driving car, the environment is the road, with all its traffic, pedestrians, and weather. For a spam filter, the environment is your inbox and the stream of incoming emails.

How does an agent interact with its environment? Through two key components: sensors and actuators.

Sensors are the tools an agent uses to perceive its environment. They're like the agent's senses. A self-driving car might use cameras, GPS, and radar to "see" where it is and what's around it. A chess-playing AI's sensor is simply the input that tells it the state of the board after the opponent's move.

Actuators are what the agent uses to act on its environment. They're the agent's hands and feet. For the self-driving car, the actuators are the components that control the steering, acceleration, and braking. For the spam filter, the actuator is the command to move an email to the spam folder or leave it in the inbox.

This creates a continuous loop. The agent perceives the environment, decides what to do, and then acts, which can change the environment. Then, it perceives the new state of the environment and the cycle begins again.

So how does an agent decide which action to take? It uses what's called an agent function. This is the internal logic or mapping that connects what an agent has sensed to the action it takes.

First, let's define what the agent is sensing. At any given moment, the agent's input from its sensors is called a percept.

The complete history of everything the agent has ever perceived is called the percept sequence. The agent function takes this entire history as input and outputs an action.

Mathematically, we can describe this relationship like this:

This formula simply says that the agent function, $f$, maps any given percept sequence, $P^*$, to a specific action, $A$. This function is the core intelligence of the agent. It's the rulebook, the strategy, or the complex neural network that determines the agent's behavior.

To make this more concrete, let's look at a couple of examples.

As you can see, the agent's design is completely dependent on its job and its environment. A spam filter's agent function is built to recognize patterns in text, while a self-driving car's function must interpret a constant stream of complex visual data to navigate the physical world safely.