The interaction of grammar and pragmatics has long been a central concern in the field of linguistics. Among the various accounts, Ariel’s “code and inference” model of grammar-pragmatics interaction is rather representative and influential. This paper aims to elucidate and critically review her model by highlighting its three key theoretical contributions: (1) relying on one single criterion—code versus inference—to distinguish grammar and pragmatics, or to delineate their division of labour; 2) integrating both diachronic and synchronic dimensions to explain grammar-pragmatics interaction; (3) proposing three innovative concepts—grammatical pragmatics, truth-compatible inference, and privileged interactional interpretation—to capture the dynamics of how grammar interacts with pragmatics. However, Ariel’s model is found to be limited as well: the dichotomy of code and inference does not fully uncover the complexity of the grammar-pragmatics interactive relationship, the strict binary distinction or opposition between code and inference is too absolute, and tends to marginalize or even disregard the conventional dimension of pragmatic analysis. Therefore, we propose to moderately modify the code-inference dichotomy and replace it with conventionality and intentionality, which we argue provides a more nuanced and comprehensive explanation of the grammar-pragmatics interaction.