Gaming is widely spreading in education. In gaming, learners make decisions iteratively in a simulation environment that replicates a real-world problem, called a game, and study what the proper decisions are in the game before making decisions in the problem. In the game, we need an expert agent that always makes proper decisions and from which the learner can learn such decisions by watching its behavior and/or by investigating it. However, it is difficult to develop such an expert agent manually because it is a very heavy task to implement knowledge of human experts on the problem into the agent, and sometimes it is also hard to extract such knowledge from the experts. Instead, we propose to automatically develop an agent that acts rightly as human experts with evolutionary computation. In particular, in this paper, we evolve such an agent for a research career design game where the player first determines his/her goal and has to decide his/her choices many times to achieve the goal afterward, based on a huge number of different situations. The experimental result shows that the evolved agent is better than or, at least, similar to agents created by a human expert.