Many software projects spend a significant proportion of their time developing the User Interface, so any degree of automation in this area has clear benefits. Research projects to date generally take one of three approaches: interactive graphical specification tools, model-based generation tools, or language-based tools. The first two have proven popular in industry but are labour intensive and error-prone. The third is more automated but has practical problems which limit its usefulness. This paper proposes applying the emerging field of software mining to perform runtime inspection of an application's architecture and reduce the labour intensive nature of interactive graphical specification tools and model-based generation tools. It also proposes UI generation can be made more practical by delimiting useful bounds to the generation process. The paper concludes with a description of a prototype project that implements these ideas.