Justifying the design of ai interventions in education: An analysis of teachers’ attitudes and a multi-layered readiness model

Introduction. The integration of artificial intelligence (AI) into general secondary education is accompanied by a persistent gap between teachers’ declared readiness to use AI tools and their actual integration into the learning process. Teachers employ AI primarily for routine tasks, while its potential for transforming instruction and personalization remains unrealized, creating risks of cognitive debt for students and the erosion of professional agency. The lack of empirically grounded models for designing AI interventions that account for teachers’ attitudes and cognitive risks impedes pedagogically sound AI integration. This study aims to empirically substantiate a two-axis ‘cognitive offloading – agency’ model and a four-layer ‘AI Readiness – 4L’ model based on an empirical analysis of teachers’ attitudes, practices, and barriers to AI use, as well as to formulate architectural principles for AI interventions. Materials and Methods. The theoretical framework comprised a synthesis of the ‘Systems AI Readiness’ concept, the SAMR, SCOT, JTBD, and TPACK models, and Bandura’s social cognitive theory. The empirical basis was a mixed-method design (2024–2025) conducted across four comprehensive secondary schools in Tomsk (Russia). The quantitative component involved a survey of teachers (n=118); the qualitative component included interviews with school principals (n=3). Thematic analysis and case study methods were employed. Results. A ‘readiness paradox’ was identified: despite positive attitudes toward AI, less than a third of teachers regularly use AI tools, and the majority rate their AI competencies as insufficient. Differentiation in principals’ managerial positions was established. A two-axis ‘cognitive offloading – agency’ model with four types of interaction in the «student – AI – teacher» system was empirically substantiated. The critical risk of cognitive debt occurs when high offloading is combined with low agency. A four-layer ‘AI Readiness – 4L’ model (technological, social, needs-based, and competency layers) was developed. Social-organizational factors, rather than infrastructure, were found to be the key predictors of readiness. Architectural principles for AI interventions were formulated. The critical risk of cognitive debt occurs when high offloading is combined with low agency. A four-layer ‘AI Readiness – 4L’ model (technological, social, needs-based, and competency layers) was developed. Social-organizational factors, rather than infrastructure, were found to be the key predictors of readiness. Architectural principles for AI interventions were formulated. Conclusions. The proposed models and principles provide a foundation for designing AI interventions that leverage the transformative potential of technology while preserving teacher agency and student cognitive autonomy, confirming the necessity of transitioning to pedagogically mediated AI adoption.