Most of the mistakes committed by users are preventable. This thesis describes the AssistU approach which iteratively interprets and reacts on the user's input during runtime. The underlying assumption is that unintentionally the user reveals more information than merely the activated function. Extending a model-based user interface development approach, AssistU implements a feedback loop. This loop collects information, analyzes according to metrics, decides according to rules, and acts to adapt the user interface.