Artificial Intelligence in Higher Education (2019–2026): A Systematic Review of Applications and an Evidence-Derived Multi-Axial Taxonomy

Generative and agentic artificial intelligence (AI) have expanded AI functions in higher education, challenging taxonomies developed before the generative-AI wave. Following PRISMA 2020, we reviewed 228 peer-reviewed journal publications from January 2019 to March 2026. The publication was the coding unit; codes represented the dominant educational AI use foregrounded in each publication. Inter-coder agreement was near-perfect for the Axis 1 coding scheme (Cohen’s κ = 0.98) and substantial overall for the observed interpretive distinctions (κ = 0.69–0.78). Full-corpus mapping placed 164 of 228 publications (71.9%) outside the four original Zawacki-Richter et al. (2019) categories, but this included transversal ethics/governance and adjacent scholarly-practice literature. The functional-adequacy test therefore used the 154 publications with a classifiable application-focused function: 64 (41.6%) fit the original four categories, whereas 90 (58.4%) required one of three additional domains—AI-enabled student support, content creation and instructional design, or institutional analytics and management. This result was essentially unchanged after removing the 2023 cohort (39/67, 58.2%), and all three added domains persisted in the held-out 2026 search block. Within functional categories, applications also differed in agency, epistemic position, and pedagogy. For agency, the corpus supports an observed assistive–augmentative–delegative gradient; the autonomous endpoint is retained as a prospective theoretical category rather than an empirically validated level. Pedagogical paradigm was less stable and remains the most provisional interpretive axis. We propose a configurational four-axis framework with ethics and equity as a transversal layer, intended to support decisions about oversight, verification, pedagogical alignment and governance rather than to predict educational outcomes.