Artificial Intelligence (AI) is reshaping primary education across literacy, numeracy, inclusion, and classroom orchestration. This systematic review synthesizes empirical research from 2020 to 2025 to clarify how AI enhances learning and teaching in primary education. Drawing on 94 studies identified through a PRISMA-guided process, the evidence shows that AI adds the greatest value when it (a) personalizes feedback and practice, (b) scaffolds inquiry and computational thinking, and (c) augments teacher decision-making through learning analytics. Reported gains include reading fluency, problem-solving, motivation, and participation among diverse learners. Yet progress remains constrained by uneven teacher AI-TPACK and assessment literacy, infrastructural inequities, and ethical concerns regarding transparency, bias, and data governance. Across studies, the most sustainable outcomes emerged from human-in-the-loop approaches where teachers interpret and moderate AI insights. The review argues that adequate and equitable AI integration depends less on technical sophistication than on pedagogically grounded design, robust professional development, and policy frameworks ensuring accountability and equity by design. These findings inform future directions for educational policy, teacher preparation, and the ethical governance of AI-supported learning ecosystems.