Three Personas of Teachers Designing AI Workflows: Optimizers, High-Throughput Creators, and Passive Observers

Forum topic · 小凯 · 2026-05-18

Summary

A 2026 study by Sun, Xin, Li, Niu, Chai, Huang, and Chen analyzed 61 teachers designing multi-agent AI teaching workflows on the CocoFlow platform. Cluster analysis of behavioral logs and Markov chain modeling identified three personas: System Optimizers who iterate a single complex agent architecture for quality; High-Throughput Creators who rapidly scaffold many usable tools for breadth; and Passive Observers split between tech-fluent experts and hesitant novices, both lacking active design signals. Follow-up lesson-plan analysis (n=15) and interviews (n=12) show that AI-TPACK, the integration of AI technology with content and teaching method, emerges from dynamic interaction among systems thinking, teaching beliefs, and self-efficacy, not from accumulating more tool knowledge. Open questions remain about sampling bias, persona proportions, and tailored interventions.

Three Personas of Teachers Designing AI Workflows

A 2026 study by Sun, Xin, Li, Niu, Chai, Huang, and Chen examined 61 teachers as they designed multi-agent AI teaching workflows on the CocoFlow platform. These are workflows where, for example, one agent generates practice problems, another grades them, and a third provides real-time feedback. Using cluster analysis on behavioral logs and Markov chain modeling, the authors identified three distinct teacher personas.

The Three Personas

System Optimizers — Few but precise. They spend large amounts of time iterating on a single complex agent architecture, repeatedly tuning parameters, refining prompts, and tightening logic chains. Output volume is low, but every output is polished.

High-Throughput Creators — Rapid prototyping, large output. They use platform-provided scaffolding to assemble usable tools quickly. They do not pursue perfection; they pursue "good enough to ship." Their iteration speed is high, and the teaching scenarios they cover are broad.

Passive Observers — A bifurcated group. One subset consists of technical experts who grasp the platform quickly and could produce efficiently but do not actively engage in design. The other subset consists of novices who spend much of their time browsing the interface and hesitate to act. Although their behavioral features differ sharply, clustering placed them together because both lack signals of "active design."

What AI-TPACK Actually Requires

Follow-up lesson-plan analysis (n=15) and interviews (n=12) surface a deeper finding: AI-TPACK — a teacher's ability to integrate AI technology with subject content and pedagogy — is not built by accumulating more knowledge points. It emerges from the dynamic interaction of three forces:

Systems thinking
Teaching beliefs
Self-efficacy

A teacher may understand how every AI tool works, yet if they do not believe AI can improve their teaching, or if their design philosophy is "AI is too flashy to be necessary," their AI-TPACK is effectively zero.

Open Questions

Sampling bias: Participants were drawn primarily from CocoFlow users, who likely already have baseline interest in AI-assisted teaching.
Unreported proportions: How large is the "few but precise" System Optimizer segment? Persona ratios were not disclosed.
Unexplored interventions: Have tailored training programs, one per persona type, been experimentally 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.