In this article, I examine some of the ideas about task-based language teaching (TBLT) which have emerged over the 17 years of the current editorship of ELTJ, focusing in particular on grammar and vocabulary, and enquiring to what degree these ideas take adequate account of classroom context. Over this period, TBLT scholars have built up a sizeable body of research and made a variety of influential recommendations for classroom practice. Underlying much of this research is the belief that we can make reasonably reliable predictions about how tasks or other forms of classroom interaction will trigger certain learning processes, without needing to enquire too deeply into matters of context (such as the social relationships between learners). Is this a fair assumption, or do we need research to take classroom contexts more seriously if it is to be beneficial to both language learners and language teachers?