Agents & Search›Core· 30 min read

Intelligent Agents

Most AI can be described as an agent: it senses its environment and chooses actions to reach a goal.

What you will learn

Define agent, environment, percept and action
Break a real AI into the agent model
See goals as what drives behaviour

The agent model

A huge amount of AI fits one simple picture. An agent is anything that senses its environment and acts on it to reach a goal.

Part	Meaning	Vacuum robot example
Environment	The world it works in	Your rooms
Percepts	What it senses	“dirt here”, “wall ahead”
Actions	What it can do	move, turn, suck
Goal	What success means	a clean floor

The agent runs a loop forever: sense → think → act → sense again.

Every agent — robot or software — follows this sense–decide–act loop

# The agent loop, in pseudo-Python
while True:
    percept = sense(environment)     # what do I see now?
    action  = decide(percept)        # the "AI" lives here
    do(action)                       # change the world

Note: Output: (There is no output — this is the shape every agent program follows. The intelligence is inside decide().)

Agents do not need a body

A spam filter is an agent too: its environment is your inbox, its percept is a new email, its actions are inbox or spam, and its goal is a clean inbox. Same model, no robot.

Tip: When you meet any AI, ask the four questions: what is its environment, what can it sense, what can it do, and what is its goal? That instantly tells you how it works.

Q. For a self-driving car, “press the brake” is an example of a…

Answer: Braking is something the agent does to the world, so it is an action. Camera input would be a percept; arriving safely is the goal.

✍️ Practice

Map a thermostat onto the agent model (environment, percepts, actions, goal).
Write the sense–decide–act loop in words for a chess AI.

🏠 Homework

Choose any AI product and fill in all four parts of the agent model for it.