Communication has never occupied such a pervasive and immediate place in human life. As AI-driven conversational agents become increasingly ubiquitous across industries and online social networks evolve, novel communicative contexts are continually taking shape across scales. The arrival of these new contexts inevitably raises questions about the robustness of existing findings, especially when considering high-stakes situations related to health, politics, and law. Accordingly, the proposed dissertation takes a step towards addressing these questions by advancing complexity theory as a meta-theoretical lens for pragmatics and sociolinguistics. The contributions of this work unfold across three studies that conceptualize context as operating across scale, both as a site of convergence and as a source of variation. Beginning with contextual effects on the order of single utterances, I present original research on the pragmatics of state-of-the-art language models, providing evidence that reasoning about context in modern AI systems exhibits fundamental instability. Moving to group-level interactions, I propose an observational study examining the sociolinguistic status of Bluesky Starter Packs. Finally, at the population level, I introduce a mathematical model that explicitly encodes the inferential processes underlying situated language use, enabling the approximation of informational constraints on communication in social networks.