The game development process involves many different disciplines, ranging from visual arts, sound design, game design, product management, frontend- and backend development and many more. All of which are contributing to a single project to create the best possible player experience. The player experience can easily be interrupted by bugs or imbalanced game design. For example if a strategy game has many types of buildings, to which one outperforms all others, it will be highly likely that players would use this advantage and adopt a strategy that may reduce the usage of the surrounding buildings. The game would become monotonous instead of diverse and exciting. This work covers exploratory research whether machine learning can be used during the process of game development by having an Artificial Intelligence agent testing the game. Previous studies from other researchers have shown that Reinforcement Learning (RL) agents are capable to play finished games on super human level. However, none of them are used to improve the games themselves as they treat the environment as unchangeable, but are still facing the problem that game environments often are not flawless. To improve the game itself, this work takes advantage of the reinforcement learning algorithms to unveil the bugs and game design errors in the environment. Within this work the novelty deep reinforcement learning algorithm Proximal Policy Optimization  will be applied to a complex mobile game in its production cycle. The machine learning agent is developed in parallel to the game. As a result, the trained agent shows positive indications that the development process can be supported by machine learning. For a conclusive evaluation the additional implementation effort needs to be taken into account right from the beginning.