As AI is increasingly integrated into real-time collaboration platforms — such as writing assistants and code-generation tools — it becomes essential to understand how humans co-create with AI. In this paper, we investigate how recursive learning and feedback loops between humans and AI systems can shape the process and outcomes of collaboration. Using interaction logs, revision history data, and user interviews across contexts such as Google Docs and GitHub Copilot, we explore how users adapt their behavior in response to AI suggestions and how AI systems adapt in turn. Results unveil mutual adaptation dynamics of the emergence of collective intelligence and human-machine coevolution. We introduce a sociotechnical feedback loop framework for conceptualizing these dynamics and sketch its implications for tool design, user training, and AI governance. The present research also enriches the emerging field of hybrid intelligence by providing empirical and theoretical evidence for recursive co-creation in digitally mediated contexts. This paper introduces an original empirical-operational approach to algorithm and human routine coevolution, using distributed feedback loops on real-time tools.