The history of CAAD research is largely one of generic computing techniques grafted on to existing design practices. The motivation behind such research, on different occasions, has been to automate some or all of the design process, to provide design assistance, to check designs for compliance against some predefined criteria, or more recently to enable people to experience designs as realistically as possible before they are built. But these goals remain unexamined, and their fulfilment is assumed to be a self-evident benefit. In the worst cases, they are examples of barely concealed technology-push. Few researchers have stated in detail what they want computers to do for architectural design, most choosing instead to focus on what computers can do, rather than what is needed. This paper considers what we want CAAD systems to do for us. However, this will be a modest effort, a beginning, a mere sketch of possible directions for CAAD. But it should open channels for criticism and serious debate about the role of CAAD in the changing professional, social and cultural contexts of its eventual use in education and practice. The paper, therefore, is not so concerned to arrive at a single'right'vision for future CAAD systems as concerned by the lack of any cogent vision for CAAD.