Agents can be classified by how they decide what to do. The main types are goal-based, reflex (simple or model-based), and utility-based. Each suits different problems.

They have a clear goal and choose actions that move them toward it. They can plan ahead and consider sequences of actions. Best when the task has a well-defined end state (e.g. "book a flight and hotel for these dates").

They react to the current situation only—no memory, no long-term plan. "If I see X, do Y." Fast and predictable. Good for narrow, repetitive tasks (e.g. rule-based routing).

Like simple reflex, but they keep an internal model of the world (or recent history). So they can handle partially observable environments and short-term context.

They choose actions that maximize a utility (score). Useful when there are many ways to reach a goal and you care about trade-offs (cost, speed, quality).