Fostering collaborative futures: Multidisciplinary approaches to AI integration in educational ecosystems

Synthesis: Majumdar, Ifenthaler, Schumacher, Donlon, Hsu, Zagami, Heitink, and Mueller (2026) examine the impact of AI on education from the perspectives of researchers, practitioners, and policymakers, adopting an ecosystem perspective on AI integration in educational ecosystems. Based on the work of a Thematic Working Group at EDUsummIT, the authors conducted a Delphi study of N = 70 international professionals from 18 countries, followed by in-depth face-to-face discussions with international experts and persona-based focus-group discussions. The three most important trends were AI-related competences for learners and educators, automated formative assessment and Feedback with AI agents, and the evolved roles of teachers to teach with AI agents; top importance also included explainability and trustworthiness of AI agents and Accessibility and equal access to AI agents.

Core Finding

A Delphi study of 70 international experts (from academia, research, industry, policy, and teaching, averaging 19.5 years of professional experience) identifies five priority trends for AI integration in educational ecosystems. AI-related competences for learners and educators ranked highest in both impact (M = 8.57) and importance (M = 8.64); automated formative Assessment and feedback (M = 8.13) and evolved teacher roles (M = 8.13) tied for second-highest impact. Explainability and trustworthiness (M = 7.84) and accessibility and equal access (M = 7.81) also ranked prominently. The evolved teacher role was rated among the most challenging themes (M = 8.40), the only negative priority indicator among top-ranked themes, signaling deep contestation around professional identity.

AI in Education: An Ecosystem Perspective

An education ecosystem is defined as a dynamic, sustainable, interconnected network of individuals, institutions, data, technologies, policies, processes, and cultures that collectively influence teaching and learning in formal, non-formal, and informal contexts. The paper argues that AI adoption in education is not solely a technological issue but a systemic process shaped by interrelationships among pedagogical practices, institutional structures, AI Governance, infrastructure, stakeholder capacities, and equity considerations. Historically, AIEd evolved from intelligent tutoring systems to generative AI tools, and systematic reviews identify five principal domains: personalized learning, AI-supported formative assessment, virtual tutors and learning companions, evolving teacher roles, and ethical governance. The study's novelty lies in its ecosystem perspective, aiming for an integrated understanding of how these dimensions influence one another.

The Five Priority Trends

AI-related competences for learners and educators was the highest-rated theme, reflecting the field's consensus that meaningful AI integration depends on human capacity and readiness. It encompasses technical skills and cognitive, ethical, and emotional dimensions — for learners, understanding AI concepts and critical thinking to evaluate AI-generated content; for educators, proficiency in using AI tools while maintaining a critical perspective on their limitations. Automated formative assessment and feedback with AI agents offers real-time, individualized feedback that supports iterative and reflective learning, but raises challenges of validity, reliability, and the sociotechnical complexity of deployment. Evolved roles of teachers positions educators as orchestrators of student-centered environments working alongside AI agents, raising questions about teacher Learner Agency and professional judgment. Explainability and trustworthiness concerns the transparency with which AI systems operate and extends to reliability, fairness, and accountability. Accessibility and equal access spans design usable by all regardless of ability, plus broader socio-economic disparities and vigilance against algorithmic bias.

Structural Tensions in AI Integration

A critically reflective reading surfaces four tensions. First, between pedagogical transformation and teacher professional identity: top-down mandates without attention to the experiential and identity-related dimensions of teaching may generate resistance, superficial compliance, or professional disengagement. Second, between the competence-building imperative and uneven structural conditions for professional learning, given disparities linked to institutional resources, geography, socioeconomic context, and teacher workload. Third, between the promise of automated assessment and its sociotechnical preconditions — automated feedback systems are trained on historically produced datasets that embed existing biases, and their outputs can differentially advantage or disadvantage learners by linguistic background, learning style, and socioeconomic context. Fourth, and most structurally pervasive, the potential for AI to exacerbate rather than ameliorate educational inequality: access is contingent on infrastructure, connectivity, device availability, and digital fluency, and without deliberate equity-oriented policy, AI diffusion may follow the well-documented pattern in which well-resourced institutions capture disproportionate benefits.

Recommendations and Policy Implications

The findings converge strongly with UNESCO's Guidance for Generative AI in Education and Research, but extend it by showing that these are empirically prioritized concerns that experts perceive as high-impact and, in several cases, highly challenging to enact. The paper organizes strategies for policymakers, researchers, and practitioners across domains including equitable access, data protection, and Privacy (e.g., design AI-enhanced ecosystems guided by principles of inclusivity and equity such as universal design for learning); a shared vision of the education ecosystem; acceptable usage of AI; fostering a culture of lifelong learning; societal understanding of AI in education; and responsible use of Multimodal AI data (e.g., research on the effects of bias in educational data and co-designing ecosystems with open data). Sustainable and ethical adoption cannot be achieved through piecemeal interventions but requires coordinated action across policy, practice, and research.

What this means for practice

• Administrators. Fund AI competence building for both learners and educators first; experts rated this the top trend for impact (M = 8.57) and importance (M = 8.64) across the whole panel.
• Administrators. Treat the shift in teacher roles as a professional-identity issue, not a training logistics issue: it was rated among the most challenging themes (M = 8.40) and was the only top-ranked trend with a negative priority indicator, so top-down mandates without attention to teachers' experience invite resistance or superficial compliance.
• Policymakers. Require explainability, reliability, fairness, and accountability as procurement conditions for AI agents (importance M = 7.84), and audit infrastructure, connectivity, device availability, and digital fluency before scaling deployments, since access gaps can let well-resourced institutions capture the benefits.
• Policymakers. Set validity and reliability requirements for automated assessment and feedback systems before adoption; the paper flags that these systems are trained on historically produced datasets that embed existing biases and can advantage or disadvantage learners by linguistic background, learning style, and socioeconomic context.
• Researchers. Co-design AI-enhanced ecosystems with the stakeholders who use them and publish on bias in educational data and open-data approaches, since the panel identified research-practice-industry coordination as part of the agenda.

Limitations

• The Delphi panel was N = 70 (36 female, 33 male, 1 not reported) recruited by convenience sampling across 18 countries; Ireland (n = 16) and Germany (n = 15) supplied nearly half the participants, so regional coverage is uneven.
• Participants came from academia (n = 26), research (n = 23), industry (n = 5), policy (n = 8), and teachers (n = 8); students and parents — the target users of the systems under discussion — were not involved at this stage.
• The authors state that the Delphi pursuit of consensus may narrow the scope of perspectives and that the persona-based discussions involved interpretive synthesis by the research team, so the findings are "a structured expert-informed synthesis rather than a direct aggregation of individual perspectives."
• Priority ratings rest on expert perception using 10-point importance, impact, and challenge scales rather than measured outcomes, and the three dimensions did not differ significantly, F(2, 42) = 0.91907, p > .05.

Citation

Majumdar, R., Ifenthaler, D., Schumacher, C., Donlon, E., Hsu, H.-P., Zagami, J., Heitink, M., & Mueller, W. (2026). Fostering collaborative futures: Multidisciplinary approaches to AI integration in educational ecosystems. Computers and Education Open.