Thought experiments are widely employed in the social sciences, as many experiments are not affordable or even impossible to execute. Informal thought experiments, which are typical of classical economics, involve vagueness in the inference from their premises. On the other hand, mathematical models lack realism in their assumptions. Agent-based models are a particular kind of thought experiments, which are especially useful when our intuition is dealing with complex phenomena. We will argue that, contrary to mathematical models, agent-based models allow more realistic assumptions. On the other hand, unlike informal thought experiments, the conclusion of a simulation is the result of a strictly deductive procedure. Hence, agent-based models improve the realism of the assumptions of mathematical models, while avoiding the vagueness of informal thought experiments.