Navigating AI's educational future: expert scenarios and implications for teaching and teacher preparation

Synthesis: Twenty educational researchers from 13 countries, convened as Thematic Working Group 5 for UNESCO's EDUsummIT 2025, co-constructed six scenarios imagining AI-enriched schooling in 2030 — primary, secondary, technical and community college, and three higher-education visions. Reflexive thematic analysis of these expert artifacts, carried out with GenAI models under human interpretive authority, coded 329 references into 12 themes and condensed them into three meta-patterns: the more-human paradox, in which AI saturation intensifies rather than displaces human pedagogical work; the agency-Scaffolding tension, in which student Learner Agency with AI must be deliberately developed rather than assumed; and contextual variation, in which the meaning of critical AI partnership takes developmentally distinct forms across primary, secondary, technical college, and higher education contexts. The authors read this as pointing beyond tool training in pre-service teacher preparation toward three priority areas: developing critical, agentic AI partnerships; orchestrating AI-enhanced learning ecosystems; and building trustworthy, equitable AI infrastructure.

Key Findings

1. The expert panel comprised 20 researchers from 13 countries and 18 institutions, working as TWG5 on AI literacy for teaching and learning under UNESCO's EDUsummIT 2025.
2. The analysis coded 329 references across the scenarios and identified 12 themes; nine AI-proposed candidate themes became twelve final themes through human disaggregation of one theme into three.
3. Three meta-patterns organize the twelve themes: the more-human paradox, the agency-scaffolding tension, and contextual variation across educational phases.
4. The more-human paradox holds that AI saturation intensifies, rather than displaces, human pedagogical work: in one scenario the teacher's role "has evolved but not diminished," and AI handles routine work so teachers can facilitate interaction and coach social skills.
5. The agency-scaffolding tension holds that students' critical, agentic partnership with AI must be actively scaffolded rather than assumed, spanning primary through higher education.
6. The authors derive three priority areas for pre-service preparation: developing critical, agentic AI partnership; orchestrating AI-enhanced learning ecosystems; and building trustworthy, equitable AI infrastructure.
7. Access, inclusion or wellbeing appeared in all six scenarios (frequency 22), while tensions, limits and unintended consequences surfaced in only three scenarios, all in higher education.

How the 2030 scenarios were built

This is a futures-thinking study, not a classroom experiment. The 20 researchers, representing 13 countries and 18 institutions, worked in sector-organized small groups under a single scenario cue: "It's 2030," their institution had been an AI literacy pioneer five years earlier, teachers were well versed in pedagogy with AI, and AI literacy was embedded throughout the curriculum. Groups visualized a typical day for teachers and leaders in a mature AI-enriched environment, then discussed how approaches to teaching might evolve and how student AI literacy would be assessed and supported.

The demographic table records 7 male and 13 female participants, with Europe (10), Australia and New Zealand (5), the USA (3), and the Middle East and North Africa (2) represented, and affiliations split between 17 educational institutions and one non-profit research and development firm. Each group drafted and iteratively refined its scenario around three focal areas: teachers' priority AI literacy needs, the critical capabilities educators would need, and systemic barriers requiring intervention. The data corpus was the six final scenario artifacts — across primary, secondary, technical and community college, and higher education, roughly 15 pages in total — treated as co-constructed expert artifacts rather than transcripts or observational data.

Twelve themes, three meta-patterns

Reflexive thematic analysis coded the scenarios into 12 themes, several of which recur across sectors. Students as critical, agentic partners with AI was the most frequent theme, carrying 44 references across five scenarios, alongside self-, co-, and socially shared regulation of learning; collaborative, communicative and socio-emotional competencies; teachers as designers, orchestrators and data-informed professionals; AI as embedded pedagogical infrastructure; and AI-enabled assessment, feedback and analytics. Two further themes — student critical AI literacy and teacher critical AI literacy — extend across five scenarios from primary school to higher education.

Three meta-patterns organize this material. The more-human paradox describes AI receding into infrastructure while human relationships move to the foreground, captured in one scenario's metaphor of AI as "the air-handling system of the learning ecosystem." The agency-scaffolding tension describes a shared expectation that students interrogate and verify AI output rather than accept it. Contextual variation describes how those shared priorities take developmentally distinct forms: primary contexts foreground socio-emotional and collaborative competencies, secondary contexts self-regulation and real-time data interpretation, technical college contexts industry partnership, and higher education a divergence between educator visions focused on pedagogy and leader visions focused on governance.

Three priority areas for teacher preparation

The paper's practical argument is that preparing teachers for 2030 requires moving beyond technical AI literacy toward what it calls AI-enhanced pedagogical knowledge. The authors map the meta-patterns onto three priority areas. Developing critical, agentic AI partnership calls for supervised practicum experiences where candidates learn to scaffold students' iterative interrogation of AI responses, examine AI output for bias and hallucination before classroom use, and decide when students should engage AI versus build unassisted capability.

Orchestrating AI-enhanced learning ecosystems asks candidates to practice running group work where AI is a collaborative partner rather than an answer source, designing protocols that keep participation equitable, and interpreting AI-generated analytics through deep knowledge of individual learners. Building trustworthy, equitable AI infrastructure asks candidates to evaluate tools for representation bias and accessibility before classroom use, understand surveillance risks, and participate in institutional governance mechanisms, from bias evaluation to algorithmic impact assessments. The authors frame institutional responsibility for procurement as sitting with institutions and policymakers, while equipping teachers to contribute to it.

AI in the analysis itself

The study's method is itself a case of Human AI Collaboration. The team followed Braun and Clarke's six-phase reflexive thematic analysis, adapted to include systematic human-AI collaboration. ChatGPT-5.1 generated initial codes and proposed nine candidate themes; Claude 4.5 Sonnet was used for theme interrogation and argumentation analysis through an "auto-interview" technique. Four human researchers validated codes and made the analytical calls, including disaggregating one overly complex theme into three.

The authors are explicit that this collaboration deepened and intensified rather than accelerated the work, and they list the lack of transparency in large language model reasoning as a limitation. The scenarios and prompt logs are published on OSF and in appendices, and the paper argues that human-AI collaboration improves analytical rigor only when researchers retain interpretive authority and hold AI contributions to the same validation standards as human analysis.

What this means for practice

• Teacher educators and faculty developers. Treat AI preparation as pedagogical rehearsal rather than tool training. The scenarios call for supervised practicum experiences in which candidates scaffold students' interrogation of AI output, examine it for bias and hallucination before classroom use, and decide when students should work with AI and when to build unassisted capability.
• Curriculum designers. Build the paper's three priority areas into programs explicitly: developing critical, agentic AI partnership; orchestrating AI-enhanced learning ecosystems; and evaluating AI infrastructure for representation bias and accessibility.
• Program leaders and administrators. The authors locate responsibility for procuring trustworthy, equitable AI with institutions, regulators and policymakers, but argue teacher candidates should still learn to take part in governance — bias evaluation, algorithmic impact assessment, and recognizing surveillance risks.
• Researchers. The scenario method is a workable template for surfacing expert assumptions about an emerging technology, but the optimistic scenario cue and the participant-researcher design are cautions to carry into any replication.

Limitations

• The participant-researcher design — the same experts who co-constructed the scenarios later contributed to their analysis — introduces a positionality risk; multi-model coding, member checking, and separating scenario authorship from the analytical sub-group mitigated but could not eliminate insider bias.
• Even when prompted to consider risks, participants produced largely optimistic scenarios with limited attention to tensions, limits, or unintended consequences; the positive "pioneer" framing of the scenario cue likely shaped this, and a neutral or critical cue may have yielded different findings.
• Large language models' internal reasoning is not transparent, so the basis on which patterns were identified or weighted cannot be fully audited; human interpretive authority was maintained throughout.
• The study is a futures visioning exercise: its scenarios are co-constructed expert artifacts and the authors do not claim predictive power, so the priority areas are analytic proposals rather than tested interventions.
• The AI-assisted analysis ran on ChatGPT-5.1 and Claude 4.5 Sonnet during in-person meetings in Ireland in 2025 and follow-up collaboration in the latter half of 2025; both the models and the AI capabilities the scenarios imagine are subject to rapid change.

Citation

Malyn-Smith, J., Torrington, J., Webb, M., Butler, D., Lee, J., Tiernan, P., Chtouki, Y., Bower, M., Yazbak Abu Ahmad, M., Carvalho, A. A., Cohen, J. D., Gorissen, P., Linden, T., MacCallum, K., Raffagheli, J. E., Røkenes, F. M., Ronci, M., Shelton, C., Smith, D., & Vartiainen, H. (2026). Navigating AI's educational future: expert scenarios and implications for teaching and teacher preparation. Computers and Education Open, 11, 100422.