The rapid development of artificial intelligence (AI) technologies has significantly influenced contemporary educational paradigms, particularly in adaptive and differentiated instruction. In 9th-grade informatics education, students demonstrate substantial differences in prior knowledge, learning pace, and cognitive readiness, whereas traditional instructional approaches typically provide uniform content. This study examines the theoretical foundations of AI-assisted differentiated instruction within the framework of Uzbekistan’s national curriculum and the Cambridge-based informatics syllabus. The research integrates Bloom’s Revised Taxonomy, Vygotsky’s Zone of Proximal Development, and Tomlinson’s differentiated instruction model with educational data mining and machine learning approaches, including Bayesian Knowledge Tracing, Item Response Theory, Deep Knowledge Tracing, clustering techniques, and Natural Language Processing. The findings indicate that AI-based diagnostic and adaptive models provide a theoretically grounded mechanism for organising A/B/C level differentiation in 9th-grade informatics. The proposed Five-Layer Content Architecture and Four-Quarterly Diagnostic Model offer a structured framework for aligning algorithmic adaptation with pedagogical objectives and national assessment standards. Overall, AI-assisted differentiated instruction is pedagogically justified and theoretically viable for informatics education in Uzbekistan, provided its implementation is harmonised with national regulations and supported by systematic teacher professional development.