Foundation
Tool literacy, prompts and first pedagogical agents
PREPARED FORUNESCODIGITAL LEARNING WEEK 2026
TEACHER AI COMPETENCIES · BOUNDARY LEARNING
From AI Tool Users
to AI Agent Creators
How teachers can create and govern pedagogical AI agents while remaining the decision-makers in AI-mediated classrooms.
cocorobo.hk/unesco2026/teacher/ ↗Haiyang Xin · CocoRobo Ltd.Paris · 2026
THE QUESTION
The question is no longer whether teachers use AI.
When AI can act, assess, guide and generate—
how do teachers remain the pedagogical decision-makers?
OUR ANSWER
Mature teacher AI competence is not using AI more often.
It is configuring when AI should act, when it should be constrained, when control should be reclaimed, and when AI should remain absent.
ACTCONSTRAINRECLAIMABSENT
FACTS · KEEP THE DENOMINATORS HONEST
One programme. Two different kinds of evidence.
1,000+teachers reachedImplementation across four Shenzhen districts + partner schools
≠
3cohorts studiedSystematic research evidence from Pingshan District
12 monthsApril 2025 → April 20264 + 1accepted conference papers + ongoing journal manuscript
A BRIDGE CONTEXT
Pingshan makes the implementation gap visible.
Advanced city-level AI momentum+Uneven district-level capacity=Teacher agency must be built, not assumed
Pingshan District teacher professional learning activity
A SUSTAINED PROGRAMME
Each cohort changed the next round.
→
Expansion
Subject communities, multi-agent workflows and seed teachers
→
Systematic
Design traces, sustained engagement and Boundary Learning
TRAIN → BUILD → STUDY → REDESIGN
NOT A ONE-OFF WORKSHOP
The unit of change was a learning infrastructure.
DISTRICT
SCHOOL
SUBJECT COMMUNITY
CLASSROOMteacher-created agents
Policy alignment · seed teachers · subject seminars · classroom cases
BOUNDARIES IN PRACTICE
Boundary Learning appeared in ordinary classroom decisions.
Scaffold exploration
AI supported diagnosis and design refinement—without producing the final answer.
Human boundary: inquiry and judgmentExpand capacity
Multi-agent workflows supported practice and feedback while teachers retained standards.
Human boundary: criteria and final judgmentProtect thinking
Teachers designed scaffolds while explicitly guarding against cognitive outsourcing.
Human boundary: student cognitive workTHE HARDNESS OF FACTS
Scale and systematic evidence answer different questions.
Can this travel?
- Four Shenzhen districts + partner schools
- District, school and subject-level formats
- Teacher-created classroom cases
What can we analyse?
- Three systematically studied Pingshan cohorts
- Platform logs, surveys and agent artifacts
- Presentations, interviews and thematic analysis
Neither layer, by itself, proves student learning gains.
THE RESEARCH PROGRAMME
Five studies form a progressive evidence chain.
Can teachers create meaningful pedagogical agents?
AERA · acceptedHow do teachers actually engage after training?
AERA · acceptedWhere is AI-TPACK visible in practice?
ISLS · acceptedWhy does creation continue—or stop?
ISLS · acceptedWhat enables mature boundary judgment?
Journal · ongoing01
FINDING · LEARNING BY MAKING
Creating agents externalizes tacit pedagogy.
Teachers turn an implicit teaching judgment into a runnable artifact—then test, revise and defend it.
DESIGN→TEST→DIAGNOSE→RECONFIGURE
02
FINDING · PARTICIPATION ECOLOGY
Training does not automatically produce active creators.
ACTIVE CREATORSbuild and revise
BROWSERSobserve and explore
LOW ENGAGEMENTrarely participate
Follow-up support, peer examples, time and community matter.
FINDINGS · DESIGN + SUSTAINABILITY
Competence becomes visible in design traces—and sustainable only with system support.
WORKFLOW CONFIGURATIONAGENT ARCHITECTURETESTING PATTERNSDESIGN REVISIONSOBSERVABLE AI-TPACK
SUSTAINED CREATION
AUTONOMYCOMPETENCERELATEDNESSSCHOOL SUPPORT
Capacity building alone is not enough.THE FRONTIER
Boundary Learning has four connected capacities.
Recognize the task boundary
Which tasks need AI—and which require struggle, dialogue or human judgment?
Diagnose the model boundary
Is failure caused by prompting, workflow, model limits or task mismatch?
Demarcate responsibility
Which decisions remain with teachers, students or peers?
Configure the boundary
Translate judgment into roles, constraints, timing and non-intervention zones.
DESIGN ETHIC
Learning needs productive friction.
AI gives complete answers too earlyDelay feedback; require an attempt first
AI-generated scripts replace student voiceRequire peer critique before AI revision
AI rubric becomes rigidReturn criteria and final judgment to teachers
AI intervention ends discussionConstrain or disable AI to restore dialogue
The goal is not a frictionless classroom. It is the right friction, in the right place.
WHAT EDUCATION SYSTEMS CAN DO
Move teacher AI policy from adoption to authorship.
- Design professional learning around creation.Teachers build, test, diagnose and revise—not just watch demonstrations.
- Assess artifacts and traces.Use workflows, agent architectures and revision histories alongside self-report.
- Embed support in schools.Protect time, autonomy, competence, peer community and follow-up.
- Make non-use a legitimate decision.Human oversight includes the authority to delay, constrain or exclude AI.
Aligned with UNESCO's human-centred AI agenda; this is a project-side mapping, not UNESCO certification.
EVIDENCE RECORD
What the evidence supports—and what it does not.
SUPPORTED
- 12-month, three-cohort programme
- 1,000+ implementation reach
- Multi-source evidence on teacher creation and participation
- Four accepted conference papers
- HKU FREC ethics approval EAE26002
NOT YET CLAIMED
- 1,000+ as a research sample
- Causal student learning gains
- Universal transfer across contexts
- Boundary Learning as a validated mature theory
- UNESCO endorsement of this project
Ongoing synthesis: Redistributing Pedagogical Control: Path-Dependent Boundary Learning…
THE CLOSING TEST
Agency,
not automation.
Can teachers decide when AI should step forward—and when it should step back?
Haiyang XinFounder & CEO · CocoRobo Ltd.tony@cocorobo.cc
Full evidence, studies and framework
cocorobo.hk/unesco2026/teacher/