Agentic and quasi-agentic AI technologies in early childhood education and care: A scoping review

Agentic AI refers to systems that pursue goals by planning multi-step actions, reasoning adaptively over data and environmental cues, and acting with limited human intervention. Such systems are reshaping K-12 and higher education, yet evidence for the youngest learners remains thin. This scoping review maps the literature on agentic and quasi-agentic AI in early childhood education and care, following Arksey and O'Malley (2005) and Levac, Colquhoun, and O'Brien (2010) and reporting in accordance with PRISMA-ScR (Tricco et al. 2018). Searches of Scopus, Web of Science, IEEE Xplore, and the ACM Digital Library, supplemented by Google Scholar, arXiv, and hand-searching, identified 744 records, of which 13 were included: five primary studies (314 preschool and early-primary children) and eight peripheral studies. Only one included system fully met this definition; the studies form a gradient from architectural through perceived to prospective agency. Pedagogical design and human relational presence proved more consequential than technological sophistication, children's anthropomorphic interpretations emerged as developmentally salient, cultural context shaped ethical deployment, and transparency about system limits supported children's critical capacities. The review identifies no longitudinal evidence, no study directly observing children under three, and a regulatory-empirical disconnect, and calls for developmentally specified research before deployment outpaces evidence.