We design and implement a hybrid agent architecture that combines emotional reactions to stimuli with reasoning about the consequences of actions as a means of developing effective behaviour in a toy world. The fundamental task of our agents is to survive by periodically finding and eating food and by dealing with hostile predators, either through flight or through flight. Agents are free to cooperate, antagonize, or ignore each other and agents with different emotional profiles will pursue different strategies. The thesis begins with theoretical investigations of different models of computations and their relation to the biological brain. Our assumption is that the brain's function is, to some degree, analogous to a collection of white boxes, observable to each other. Accordingly, our agents are modelled as a collection of loosely coupled components which communicate with each other through messages. Any component is free to read any message and components have no information about which other components read the messages which they insert into the agent's message space. The affective evaluation of its environment forms the basis of each agent's decision-making, though it is complemented by a belief generator which makes inferences about future world-states resulting from certain choices. Though evaluation of these future states as well, agents can optimize their behaviour, as they can foresee likely positive or negative consequences of their actions. The thesis ends with an evaluation of individual behaviour as well as a population-based evaluation: we evaluate our agents qualitatively by placing agents in a number of simple scenarios and observing whether they perform tasks like collecting food or avoiding predators. After that, we evaluate them in a population-based manner by placing various populations into larger scenarios and recording the survival of different personality types over time.