Despite the promise of generative artificial intelligence (GenAI) to provide individualized learning support at scale, many GenAI-powered learning systems remain weakly connected to principles from the learning sciences. When GenAI is incorporated into instruction with limited theoretical grounding, it can lead to shallow personalization that optimizes for fluency, but yields limited long-term retention and transfer. It can also create fragmented learning experiences that increase cognitive load. In addition, the use of unguided GenAI may shape classroom practices in ways that unintentionally weaken teachers’ instructional judgment and students’ self-regulated learning. This chapter argues that realizing the educational value of GenAI requires rethinking how foundational learning principles guide both pedagogy and the design of GenAI-enhanced learning environments. We revisit principles such as contiguity effects, generative activity, feedback, and self-regulated learning, and we reinterpret them for settings where content, representations, and feedback can be dynamically generated. Building on this conceptual analysis, we propose design-oriented recommendations that map learning principles to concrete GenAI affordances at the levels of learner activity, teacher orchestration, and system behavior. We conclude with illustrative design patterns, evaluation priorities that capture both learning outcomes and learning processes, and research directions that align GenAI innovation with theory-driven and ethically responsible learning design.