Computer input is more complex than a sequence of single mouse clicks and keyboard presses. We introduce a novel method to identify and represent the user interactions and build a system which predicts - in real-time - the action a user is most likely going to take next. For this, a recurrent neural network (RNN) is trained on a person's usage of the computer. We demonstrate that it is enough to train the RNN on a user's activity over approximately a week to achieve an accuracy of 34.63 % when predicting the next action from a set of almost 500 possible actions. Specific examples for how these predictions may be leveraged to build tools for improving and speeding up workflows of computer users are discussed.