The Architecture of Roles in AI-Designed Group Activities: A comparative inductive analysis of novice and experienced teachers' differentiated instruction within the IAT framework

Synthesis: The architecture of roles in AI-designed group activities. A comparative qualitative study by Hossein Talebzadeh (2026) analyzing how 89 Iranian teachers (38 novice, 51 experienced) assigned roles in 178 AI-designed group activities within the Integrated AI Triad (IAT) framework. Using reflexive thematic analysis, the study introduces three constructs — Role Richness, Role Synergy, and Level-Role Alignment — and finds experienced teachers produced significantly richer, more interdependent, and more precisely ZPD-aligned role architectures, while novices often collapsed integrated projects into conventional assessments lacking product integration. The paper introduces "pedagogical prompt literacy" as a portable construct: AI output quality is fundamentally contingent on teachers' capacity to encode pedagogical intentions through Pedagogical Content Knowledge.

Overview

Designing effective cooperative group activities that differentiate by student readiness is cognitively demanding. This study asks whether teachers' career stage shapes how they use generative AI to design differentiated group activities — and why. Within the IAT framework, teachers are positioned as "Bilingual Learning Designers" who must encode pedagogical intentions into AI prompts.

Method

• Sample: 89 Iranian teachers (38 novice, 51 experienced) across four cohorts.
• Corpus: 178 AI-designed group activities within the IAT framework (two activity models: Integrated Multi-Level and Mixed-Ability).
• Analysis: Reflexive thematic analysis with hybrid deductive-inductive coding, double-coded with intercoder reliability; trustworthiness and positionality addressed.
• Novel constructs: Role Richness (number/variety of roles), Role Synergy (interdependence among roles), Level-Role Alignment (match between task demands and student readiness/ZPD).

Key findings

• Experienced teachers design richer role architectures. 85% of experienced designs had detailed role descriptions (43/51) versus 30% of novice designs (11/38); 78% vs. 30% had 3+ distinct role types. (Percentages reflect relative thematic prevalence, not inferential claims.)
• The largest gap is Role Synergy. 90% of experienced designs showed role interdependence (46/51) versus 35% of novice designs; novices frequently produced parallel tasks completable in isolation.
• Precise Level-Role Alignment. Experienced teachers calibrated tasks to readiness — 88% had appropriately challenging advanced tasks and 90% appropriate basic tasks, with 65% explicit ZPD-calibration versus 15% for novices. Novices often used quantitative differentiation (more questions) rather than qualitative variation in cognitive demand.
• The novice–expert paradox. Experienced teachers, despite less AI familiarity, produced superior designs — evidence that pedagogical expertise, not technical proficiency, drives effective AI integration. "Smart drunk intern" framing: AI output needs constant expert evaluation.
• Pedagogical prompt literacy. Experienced teachers encoded PCK into prompts (e.g., the R.A.F. — Role + Audience + Format — framework) with greater precision. AI amplifies teacher judgment rather than replacing it.
• Experience alone is insufficient. A negative-case analysis showed one experienced teacher (19 years) produced low-synergy, non-integrated roles — experience must be paired with professional learning on integrating differentiated contributions.

What this means for practice

• Instructors. Score your AI-designed group activities on Role Richness, Role Synergy, and Level-Role Alignment before using them. Only 35% of novice designs showed role interdependence against 90% of experienced teachers' designs, and novices frequently produced parallel tasks completable in isolation.
• Instructors. Specify what each role must produce and why that role fits that student, rather than adding more questions: novices defaulted to quantitative differentiation where experienced teachers varied the cognitive demand and made ZPD calibration explicit (65% vs 15%).
• Faculty developers. Build professional learning around pedagogical reasoning and prompt specification — for example the Role + Audience + Format heuristic — not tool training: experienced teachers designed superior role architectures despite less AI familiarity.
• Faculty developers. Target experienced teachers who have not integrated differentiated contributions: a negative-case analysis found a teacher with 19 years of experience producing low-synergy, non-integrated roles, so seniority alone is not the lever.
• Researchers. Treat the three constructs and pedagogical prompt literacy as analytical tools to adapt, test, and refine in other institutional settings; the authors offer them as a replicable method rather than a claim to universal generalization.

Limitations

• Participants were 89 Iranian teachers (38 novice, 51 experienced) across four cohorts, and the authors state that the cultural and institutional context limits generalizability to other educational systems.
• Data are 178 activity designs produced in an IAT workshop: they reflect design intentions, not classroom implementation, so the study does not show whether the designs were used or influenced student learning.
• A single AI tool (ChatGPT-4) and specific prompt templates were used, so findings may not generalize to other generative AI systems.
• The researcher was both framework developer and analyst, creating potential confirmation bias despite mitigation, and self-report elements may involve social desirability; double-coding covered ~20% of designs (overall Cohen's κ = 0.82).

Citation

Talebzadeh, H. (2026). The Architecture of Roles in AI-Designed Group Activities: A comparative inductive analysis of novice and experienced teachers' differentiated instruction within the IAT framework. Farhangian University, Tehran, Iran.