Abstarct: In the software development environment, especially with the new agile methods and its popular subsets such as CI/CD, software testing is vital. For the types of test levels in a software, several test cases are defined, all of which must be performed after each development or error detection cycle and before publishing, to further guarantee the health and functionality of the software and, as a result, user satisfaction. to be Several methods have been presented to solve this problem until now, in two very general cases, it is done by a human expert, and the second case is the training and use of an intelligent agent that can make decisions with the lowest error rate in this environment. In the previous solutions, observational and reinforcement learning methods were used to solve the problem, which led to the production of factors with appropriate accuracy. In continuation of this path, the future research also considers this issue as an environment for multi-agent game, with the reinforcement learning method, the result of which is the improvement of the results with an acceptable difference compared to the previous methods. The aim of this research is to achieve intelligent agents using multi-agent reinforcement learning. Agents who have learned well and prioritize the test items with the desired accuracy and speed. Prioritization means that the test cases that, if we change their position in the execution queue and are executed earlier, errors will be discovered faster. As a result, the development and troubleshooting phase is done faster and the quality product reaches the consumption cycle.