Yimeng Sun, Haiyang Xin, Shuang Li, Qiannan Niu, Ching Sing Chai, Lingyun Huang, Gaowei Chen (2026) — Multiple institutions. arXiv:2605.13906 [cs.CY].
Key Findings
This study investigates how teachers design multi-agent instructional workflows and identifies three distinct teacher archetypes that emerge from behavioral log analysis of 61 in-service teachers:
- Systematic Optimizers: Iteratively refine complex multi-agent architectures with a methodical, architectural focus. These teachers treat AI agent orchestration as a systems design problem.
- Prolific Creators: Rapidly prototype pragmatic tools, leveraging scaffolding to quickly produce usable instructional aids. Efficiency-oriented and output-focused.
- Passive Observers: Exhibit polarized expert-novice profiles with minimal proactive modification of the AI workflows presented to them.
AI-TPACK Beyond Discrete Knowledge
The core theoretical contribution is that effective AI-TPACK integration — the fusion of Technological, Pedagogical, and Content Knowledge in an AI context — emerges not from possessing separate knowledge domains, but from a dynamic interplay of three factors:
1. Systems thinking: The ability to conceptualize interactions among multiple AI agents and map them to instructional goals 2. Pedagogical beliefs: Underlying assumptions about teaching and learning that fundamentally shape design choices 3. Self-efficacy: Confidence in one's ability to orchestrate complex AI-enhanced workflows
This challenges static, checklist-based models of teacher-ai-competency and aligns with broader research showing that faculty-development-genai must address cognitive-behavioral diversity, not just technology training.
Implications for Teacher Professional Development
The findings call for differentiated scaffolding:
- Systematic Optimizers benefit from advanced system-design frameworks and revision opportunities
- Prolific Creators thrive with rapid-feedback cycles and modular, reusable components
- Passive Observers need explicit modeling, guided practice, and confidence-building exercises
This connects to the teacher-ai-adoption-confidence finding that teacher confidence fully mediates institutional support effects on AI adoption. It also extends the teachingcoach-chatbot-instructor-guidance paradigm by suggesting that coaching scaffolds must be personalized to teacher archetypes, not just content domains.
Connection to Agentic AI in Education
The multi-agent workflow framing positions this work at the intersection of agentic-ai-ecosystems-higher-education and teacher professional learning. As eduagentbench-agent-teaching-benchmark establishes benchmarks for what agent tutors should do, this study addresses the complementary question: how should teachers learn to design and orchestrate those agents?
Related Pages
- cyberscholar-genai-writing-feedback — Generative AI Feedback, English Writing and Teacher Rubrics: A Multiple-Case Study of CyberScholar
- ai-tutor-authoring-promptdecipher — PromptDecipher: Supporting AI Tutor Authoring Through Editable Simulated Interactions
- teacher-ai-adoption-confidence — Teacher confidence mediates institutional support for AI adoption
- faculty-development-genai — Approaches to developing faculty capability with generative AI
- teachingcoach-chatbot-instructor-guidance — Fine-tuned scaffolding chatbot for instructors
- agentic-ai-ecosystems-higher-education — Multi-agent AI frameworks for higher education
- eduagentbench-agent-teaching-benchmark — Benchmark for holistic AI tutor evaluation
- ai-generated-slides-student-perception — Student perception of AI-generated educational content
- teacher-ai-competency — Frameworks for teacher AI competency development