From information providers to learning designers: heutagogical teaching practices in AI-enhanced education

Synthesis: Hochberg and Waitzman ask how teachers translate AI-focused professional learning into designs that give learners real control. Their qualitatively driven embedded mixed-methods study followed 93 Israeli teachers through an eight-week professional development program on AI, self-regulated learning, and learner-centered pedagogy. The primary qualitative strand analyzed more than 280 units — reflections, open-ended responses, lesson plans, and instructional artifacts — through hybrid deductive-inductive thematic analysis, while an exploratory quantitative strand compared 61 pre-program with 26 post-program questionnaire submissions using Welch tests. The analysis produced five interconnected dimensions: designing for learner agency, self-regulated teaching and reflection, AI-supported learning design, designing for capability development, and critical AI literacy. Post-program means were descriptively higher for reflective teaching, AI-supported planning, and inquiry-oriented design, but none of the four comparisons was significant after Holm adjustment. The authors propose a dual-layer relationship in which teachers' own reflective regulation and critical AI Literacy mediate designs intended to support learner Learner Agency and capability; because the study measured teachers' reports and artifacts, its contribution concerns pedagogical intentions and professional learning, not demonstrated gains in learner agency.

Key Findings

1. Teachers described moving from answer-oriented instruction toward activities in which learners formulate questions, compare explanations, discuss alternatives, and participate in inquiry — intent to design for learner Learner Agency.
2. Generative AI entered planning as a provisional thinking partner: participants generated and revised lesson ideas, prompts, assessment tasks, sequences, pacing options, and alternative pathways, then adapted the outputs rather than copying them.
3. Teachers used AI to analyze lessons, generate reflective questions, identify difficulties, and revise subsequent plans — a planning, monitoring, evaluation, and revision cycle read as teacher-level self-regulation.
4. Critical AI literacy emerged as both a learning-design goal and a condition of teachers' own mediation, with participants asking learners to detect inaccuracies, contradictions, bias, and misinformation in AI outputs.
5. Exploratory pre-post comparisons showed descriptive gains for reflective teaching (mean difference 0.40), AI-supported planning (0.15), and inquiry-oriented design (0.24), and no change for differentiated instruction (−0.01).
6. None of the four comparisons was statistically significant after Holm adjustment and every confidence interval included zero, so the questionnaire strand offers only limited directional corroboration.

What the program involved

Ninety-three teachers in the Israeli education system volunteered for an eight-week professional development program combining weekly two-hour synchronous online sessions with asynchronous tasks. They experimented with ChatGPT, Gemini, Perplexity, Canva AI, and education-specific applications while working on AI-supported planning, learner agency, self-regulated learning, reflective practice, differentiation, inquiry, and critical and ethical AI use, with emphasis on pedagogical judgment rather than tool proficiency. Before the program, 24% reported daily or near-daily AI use and 38% weekly use.

Five dimensions of AI-enhanced heutagogical teaching

The hybrid analysis generated five interconnected dimensions. Designing for learner agency appeared as question formulation, inquiry, comparison of explanations, and collaborative exploration in place of answer-oriented instruction. Self-regulated teaching and reflection appeared in AI-assisted lesson review and revision cycles — planning, monitoring, evaluation, and revision treated as teacher-level regulation rather than measured student regulation. AI-supported learning design appeared as iterative adaptation of plans, prompts, Assessment tasks, pacing, and alternative pathways. Designing for capability development appeared as dilemmas, unfamiliar scenarios, enrichment tasks, and alternative formats intended to support Transfer of Learning beyond familiar examples. Critical AI literacy surfaced as both a design goal and a condition of professional mediation.

What the pre-post comparison did and did not show

An exploratory quantitative strand compared 61 consenting pre-program with 26 consenting post-program questionnaire submissions. Because submissions were anonymous and could not be linked across time, group means were compared with two-sided Welch independent-samples tests rather than matched observations. Reflective teaching was descriptively higher after the program (M = 3.27, SD = 0.92) than before it (M = 2.87, SD = 0.98), a mean difference of 0.40 with Hedges' g = 0.41. Smaller differences appeared for AI-supported planning (0.15) and inquiry-oriented design (0.24), while differentiated instruction was essentially unchanged (−0.01). The unadjusted comparison for reflective teaching was p = 0.077 and the Holm-adjusted value p = 0.307; none of the four comparisons was significant after adjustment, and every confidence interval included zero.

Why heutagogy, and not just AI-supported pedagogy

The authors treat heutagogy as an analytic criterion rather than a synonym for technology integration or learner-centered instruction. Technological Pedagogical Content Knowledge (TPACK) can explain whether teachers coordinate technology, pedagogy, and content, but not whether learners influence purposes and pathways or develop capability for unfamiliar situations; teacher-agency theory explains professional choice, but not the criteria by which choices become self-determined designs. The proposed contribution is a dual-layer relationship: at the teacher layer, self-regulated reflection and critical AI literacy preserve professional judgment in interaction with probabilistic systems; at the design layer, that judgment informs opportunities intended to support learner agency and capability. AI-supported design connects the layers but does not guarantee learner self-determination, and the relationship is offered as a proposition for future testing rather than a demonstrated student outcome.

What this means for practice

• Instructors. Decide explicitly where control sits: when AI generates alternatives, keep the choice of questions, pathways, and verification with learners, and treat outputs as provisional drafts requiring judgment.
• Faculty developers. Build professional learning around reflective cycles and design autonomy instead of tool demonstrations; acceptance and ease of use are entry conditions, not the outcome, and assessment should draw on artifacts and rationales rather than confidence surveys.
• Curriculum designers. Build capability-oriented tasks — dilemmas, unfamiliar scenarios, multiple explanations, alternative formats — and embed critical AI literacy in planning, differentiation, inquiry, and assessment rather than a separate ethics session.
• Researchers. Test the dual-layer proposition: whether teacher reflection and critical AI literacy predict the quality of agency- and capability-oriented designs, and whether learners experience those designs as meaningful opportunities for self-determination.

Limitations

• The evidence rests on self-report and course-produced artifacts. Social desirability, demand characteristics, and awareness of the program's aims may have shaped responses, and artifacts show planned practice rather than classroom enactment.
• The questionnaire strand used anonymous, unlinked submissions (61 pre-program, 26 post-program) with unequal group sizes, author-developed exploratory indicators rather than validated scales, and a single item for differentiated instruction.
• The design had no control group, random assignment, delayed follow-up, classroom observation, or learner outcome measure, so the quantitative results cannot support causal claims.
• The sample was self-selected from one professional development program in Israel, and immediate post-program measurement does not show durability.

Citation

Hochberg, N., & Waitzman, R. (2026). From information providers to learning designers: heutagogical teaching practices in AI-enhanced education. Frontiers in Education, 11, 1905833.