What are the prospects for applying the methods of Artificial Intelligence to architectural designers' knowing-in-action? David Marr (Marr, 1982) has advanced the idea of a computational theory of vision, which requires defining the information processing tasks carried out in vision. I ask, by analogy: what are the information processing tasks carried out in design? In order to answer this question, I propose, one ought to study design phenomenology. I illustrate several such studies, based on observations of a design studio, the performance of a design exercise, and the playing of a design game. In order to simulate the transactions with a design situation illustrated in these studies, the computer would have to carry out processes that begin prior to the presentation of what are normally defined as design inputs. Such processes involve the construction of design worlds, and they include: the simplest unit of design experimentation, the designer's seeing-moving-seeing; constructing figures from marks on a page; appreciating design qualities; setting design intentions and problems; recognizing the unintended consequences of move experiments; storing and deploying prototypes; and communicating across divergent design worlds. I conclude that the practitioners of Artificial Intelligence in design would do better to aim at producing design assistants rather than knowledge systems phenomenologically equivalent to those of designers.