Because of limited cognitive resources, humans use heuristics in problem solving and decision-making. We argue that reasoning about mental states also comprises the use of heuristics. This has been tested in sequential games, in which one player’s outcomes depend on another player’s decisions. Empirical findings show that participants apply simple strategies, or heuristics, for as long as these yield expected outcomes. However, participants were able to revise their strategies when presented with superficially similar but more difficult games. We have built a flexible and task-general computational cognitive model that can simulate these findings. The model uses a heuristic for as long as the heuristic yields expected outcomes. If the model’s decisions yield suboptimal outcomes, the model updates its strategy level. This updating can be considered a deliberate process, as it is based on an interaction between factual knowledge and problem solving skills.