The field of artificial intelligence has come a long way in the past 50 years, and studies of its methods soon expanded to a field in which they are of great practical value -- computer games. The concept of intelligent agents provides a much needed theoretical background for the comparison of various different approaches to intelligent, rational behaviour of computer-controlled characters in games. By combining rationality with certain limitations of the capabilities of our agents, we can achieve very human-like behaviour. In this diploma thesis we introduced and compared various types of agents that are used in games (but not only in games) and showed how to implement meaningful, reasonable limitations to agent capabilities into the game world. The aim of this thesis is to show the strengths and weaknesses of each type of agent and decide what types of tasks it is suitable for. We showed that even the simplest agents can succeed in their tasks in certain task environments, while more difficult task environments often require a more advanced agent architecture. The addition of goals into the agent architecture had the biggest impact on the agent's behaviour, while the finite-state machine approach kept our implementation simple and compact.