Ironic descriptions subvert the norms of descriptive language. Norms have highly salient exemplars – shared stereotypes – on which speakers can draw to create a vivid description, but ironic speakers instead construct their own counter-examples, often identifying exceptional cases where the standard inferences do not hold. One can thus hone one’s facility for irony by studying the ironic descriptions of others. Indeed, specific tactics for implementing a particular strategy for irony can be acquired by observing how others use words to subvert our own expectations. This paper provides the computational foundations for uniting these ideas into a single analytical framework. These foundations comprise: a nuanced knowledge representation of stereotypes and their most salient properties, acquired from a large-scale analysis of web similes; a set of non-literal query operators for retrieving phrases with ironic potential from a large corpus of linguistic readymades (such as the Google n-grams); a corpus of annotated similes, harvested from the web; tools for detecting irony in similes harvested from the web; and automatic tools for deriving specific tactics for irony from these attested cases.