This chapter, which uses the mirror of shape grammar implementation to reflect on why shape grammars, or, more precisely, visual calculating, isn’t more popular among computational designers, proposes that visual calculating might be better understood as a tool for thinking about design, rather than as a method for designers. The chapter compares visual, symbolic and subsymbolic calculating in architectural design, concluding that, while specific shape grammars resemble symbolic artificial intelligence and foreshadowed today’s use of computational design for design automation, visual calculating appears to evade the distinction between symbolic and subsymbolic artificial intelligence methods as a less popular, third alternative. The chapter then briefly summarizes the current state of shape grammar implementation, concluding that visual computing’s lack of popularity can no longer be blamed only on an absence of powerful, usable, and faithful enough shape grammar interpreters. The chapter discusses subshape recognition as a fundamental barrier to an ideal shape grammar interpreter that allows designers to use any rules they want, whenever they want. This barrier is not so much that subshape recognition is NP-hard—which could potentially be overcome by advances in computing—but that the number of subshapes is unbounded in the most general case. As such, shape grammar implementation becomes tractable only through human judgments that limit the exponential increase of possible rule applications. This principled resistance of visual computing to computer implementation —which arises from the requirement that designers should be able to do anything they want to—demonstrates that visual computing more convincingly serves as a striking reminder of the limits of artificial intelligence than as a design method. This conclusion leads to a remarkable convergence with Rittel, who identified “Sollsetzung,” i.e., arbitrary judgment, as the limit of artificial intelligence in design. Finally, the chapter proposes that, over the next fifty years, shape computation could ask what more we can learn from visual calculating about design, and how we might use contemporary artificial intelligence methods to support further advancements in shape grammar implementation.