Intuitively, diverse teams should have a performance advantage since they bring a broader range of cognitive resources to a task. However, intuition does not tell us how big that advantage might be, and how it can best be harnessed. Here, agent-based models can come in to help. By replacing the complexities of human behaviour with the behaviour of simulated agents following simple rules, they allow us to understand how various hypotheses might play out in the real world, and thus allow us to hone our intuitions and shape real-world research – yet they are still under-used in organisational and social psychology. In this talk, I present my replication of the foundational model by Hong & Page (2004), which suggested that “diversity trumps ability” and argue that the result can inform our understanding of real-world problem-solving, despite initially convincing results to the contrary presented by Grim et al. (2019). I will then present an original model that uses a genetic algorithm to represent the problem-solving process, and show how these models help to explore how to make problem-solving activities in diverse teams work. For instance, I will suggest that it is important to slow down convergence to maintain diversity.