Practice Without Pressure - Using a Teacher-Trained AI Agent to Build Exam Writing Skills in a Year 12 Girls' VCE Visual Arts Classroom

Sep 20, 2026

Uploaded by Rowan

Over many years of senior art teaching, a persistent mismatch appeared between sophisticated studio thinking and shallow or generic written exam responses. This action research study examined how a teacher-trained AI feedback agent, “Agent Art,” supported ten Year 12 girls at Presbyterian Ladies' College, Melbourne, in independent exam preparation and analytical writing. Agent Art was trained with subject content, rubrics, and VCAA assessment criteria language, using mixed-methods qualitative-led analysis across logs, surveys, focus group work, practice artefacts, and a reflexive journal. Four themes were identified, showing relational trust, low-stakes rehearsal, teaching reshaped by logs, and disengagement when feedback was generic or inaccurate, underscoring visible oversight and critical feedback literacy.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages25

Reading Time1 h 3 min

AccessFree to download

Abstract

Introduction

Intervention: Agent Art

Research Question

Literature Review

Glossary

Practice Without Pressure: Using a Teacher-Trained AI Agent to Build Exam Writing Skills in a Year 12 Girls' VCE Visual Arts Classroom Perri Winter Presbyterian Ladies' College, Melbourne, Australia Abstract Over many years of teaching senior art, I noticed a persistent gap: students demonstrating genuine artistic sophistication in studio work were frequently unable to reproduce the same depth of thinking in written exam responses. This action research study investigated how a teacher-trained AI feedback agent,“Agent Art,” shaped ten Year 12 girls' independent exam preparation and analytical writing at Presbyterian Ladies' College (PLC) in Melbourne. I designed Agent Art as an exam practice partner trained on subject-specific content, rubrics, and Victorian Curriculum and Assessment Authority (VCAA) assessment criteria language. Using a mixed-methods, qualitative-led approach encompassing AI conversation logs, preand post-intervention surveys, a focus group, individual conversations, practice exam artefacts, and a reflexive researcher journal, I applied polyangulation and analysed data inductively through open, axial, and selective coding. I identified four themes as answers to the research question: girls engaged with Agent Art because they trusted the teacher behind it; they used it as a low-stakes space for checking and rehearsing rather than for high-stakes judgement; reading the conversation logs reshaped my teaching and in turn how girls prepared; and when the tool produced generic or inaccurate output, girls disengaged unless they had been taught to evaluate feedback critically. The findings suggest that a teachertrained AI agent can meaningfully support girls' analytical writing when implemented with visible teacher oversight, relational trust, and explicit teaching of critical feedback literacy. Glossary Polyangulation: A qualitative enrichment strategy in action research in which data from multiple sources are cross-referenced to deepen understanding, surface contradictions, and strengthen the trustworthiness of interpretations, rather than to prove results statistically (Mertler, 2024) . Member Checking: A trustworthiness strategy in which the researcher shares preliminary interpretations with participants to verify that their perspectives have been accurately represented. In this study, member checking involved returning emerging themes to students in individual conversations and in the focus group. Critical Feedback Literacy: The capacity to evaluate, interrogate, and selectively act on feedback received, including feedback generated by automated or AI systems. Critical feedback literacy involves recognising both the value and the limitations of a feedback source and making informed decisions about which guidance to trust and apply. Practice Without Pressure: Using a Teacher-Trained AI Agent to Build Exam Writing Skills in a Year 12 Girls' VCE Visual Arts Classroom There is a particular frustration many visual arts teachers will recognise. A student sits down at her folio and talks about her work with fluency, insight, and genuine feeling. Then she picks up a pen in an exam and writes something generic, rote learned, or half the length it needs to be. Over many years of teaching, I have watched this gap between girls' artistic intelligence and their written exam performance repeat itself, not because they lacked knowledge, but because they lacked the practice, confidence, and space to rehearse the specific analytical writing the exam demands. The above challenge has been compounded by the social dynamics of a girls'classroom. Many of my students were reluctant to ask what they perceived as a silly question in front of their peers. This pattern is well documented: adolescent girls frequently exhibit perfectionism, a reluctance to take intellectual risks, and a heightened awareness of social judgement when seeking help (Booth & Nolen, 2012; Matthews, 2020) . The girls were more comfortable seek

FAQs

What problem does the study address in exam writing?

The study targets a recurring gap where students show strong artistic thinking in studio work but produce generic, rote-learned, or insufficient written responses in exams.

How was the AI feedback agent (“Agent Art”) designed and trained?

Agent Art was built in-house using Google’s Gemini LLM and configured with VCE Visual Arts content, VCAA rubrics, sample responses, and the teacher’s pedagogical knowledge, functioning as a scaffold for analytical thinking.

What determines whether students engage effectively with the AI feedback?

Girls engaged when they trusted the teacher behind the agent, used it for low-stakes practice, and benefited from log-informed changes to teaching; they disengaged when outputs were generic or inaccurate unless taught to evaluate feedback critically.

Sep 20, 2026

Uploaded by Rowan