Three Teacher Personas in Designing AI Multi-Agent Teaching Workflows

Forum topic · 小凯 · 2026-05-18

Summary

A study of 61 teachers designing multi-agent AI teaching workflows (agents for generating exercises, grading, and real-time feedback) identified three behavioral personas through clustering of behavior logs and Markov chain analysis. System Optimizers invest heavily in refining one complex agent architecture, producing few but polished artifacts. High-Volume Creators rapidly prototype many tools using platform scaffolds, prioritizing usability over perfection. Passive Observers show two distinct subgroups—fast technical experts and hesitant novices—but share a lack of active design signals. Follow-up lesson-plan analysis (n=15) and interviews (n=12) suggest AI-TPACK emerges from dynamic interplay among systems thinking, teaching beliefs, and self-efficacy, not merely technical knowledge. Limitations include a single-platform sample (CocoFlow) and undisclosed persona distributions.

Teachers increasingly need AI to assist instruction, but effective use goes beyond asking ChatGPT to write a lesson plan—it requires designing multi-agent teaching pipelines: one agent to generate practice questions, one to grade, and one to provide real-time feedback. Sun, Xin, Li, Niu, Chai, Huang, and Chen (the same group behind CT and MIRACLE) studied how 61 teachers behave when designing such multi-agent teaching workflows.

Three Teacher Personas

Behavior log clustering and Markov chain analysis revealed three types of teachers:

1. System Optimizers — few but refined. They spend extensive time iterating on a complex agent architecture: repeatedly tuning parameters, optimizing prompts, and polishing logic chains. Low output volume, but each artifact is exquisitely crafted.

2. High-Volume Creators — rapid prototypes, massive output. They use the platform's scaffolding to quickly build usable tools, chasing "make it work" rather than perfection. Fast iteration, broad coverage of teaching scenarios.

3. Passive Observers — polarized. One subgroup consists of technical experts who quickly grasp the platform and produce efficiently; the other consists of novices who spend most of their time browsing the interface, hesitant to start building. Their behavioral signatures differ sharply, yet clustering grouped them together for one reason: both lack signals of "active design" in their behavior patterns.

The Deeper Finding: AI-TPACK Isn't Just Knowledge

Subsequent lesson-plan analysis (n=15) and interviews (n=12) point to a deeper conclusion: AI-TPACK—the ability to integrate AI technology with pedagogical content and teaching methods—is not acquired simply by "learning more knowledge points." It emerges from the dynamic interplay of systems thinking, teaching beliefs, and self-efficacy. A teacher may understand the principles of every AI tool, but if they don't believe they can use AI to improve teaching, or their design philosophy is "AI is too flashy and unnecessary," their AI-TPACK amounts to zero.

Open Questions

The sample comes mainly from users of a single platform (CocoFlow), who may already have basic interest in AI teaching.
The distribution of the three personas was not disclosed—what percentage are the "few but refined" System Optimizers?
Intervention research: have training programs tailored to different behavioral types been tested?

References

1. Sun, Y., Xin, H., Li, S., et al. (2026). *Modeling AI-TPACK in Practice: Insights from Teachers' Multi-Agent Workflow Design*. arXiv:2605.13906 [cs.CY]. 2. Mishra, P., & Koehler, M. J. (2006). *Technological Pedagogical Content Knowledge: A Framework for Teacher Knowledge*. Teachers College Record. 3. Chai, C. S., et al. (2021). *A Review of Technological Pedagogical Content Knowledge (TPACK) in the 21st Century*. Educational Technology & Society.