As the complexity and scope of game development increase,  playtesting (game testing) remains an essential activity to ensure the quality of video games. Yet, the manual, ad-hoc nature of game testing gives space for improvements in the process. In this thesis, we research, design, and implement an approach to enhance game testing to balance video games. Instead of manually testing games, we present an automated approach with autonomous agents to aid game developers to assess the game's balance. We describe the process of training the agents, playing the game, and assessing the game balance using game attributes. We validated our testing process with two platform games. We conclude that the use of autonomous agents to test games is faster than the manual feedback loop and provides a viable solution for game balancing, showing spikes in difficulty between game versions and issues with the game design.