Methods and systems for selecting, testing, and applying application configurations are presented. In one embodiment, a method is provided that includes executing an application according to a first configuration and measuring a first plurality of metrics. A change to a setting of the first configuration may be identified by a machine learning model to generate a second configuration. The application may be executed according to the second configuration and a second plurality of metrics may be measured. A selected configuration for future executions of the application may be selected from among the first and second configurations based on the first plurality of metrics and the second plurality of metrics.