Innovation in Language Learning
Scaffolding Without Outsourcing: Designing a Constraint-Based AI Agent for L2 Writing
Edward Serhatti, Ufuk University (Turkey)
Abstract
As generative artificial intelligence increasingly automates student writing, a central pedagogical challenge is integrating AI without outsourcing learner cognition to the technology. This experience-based paper presents the design and classroom piloting of an AI intervention based on a pedagogical principle: AI should scaffold the writing process without producing the writing. Implemented with 30 intermediate (B1) university preparatory students, a custom AI agent guided learners through brainstorming, outlining, drafting, and revision without outputting text, paraphrasing, or providing corrections. Its Socratic model prompts students to justify choices, clarify meaning, generate alternatives, and evaluate decisions rather than receive automated answers. By preserving learner-produced text verbatim during drafting, the agent ensures students remain responsible for identifying and addressing issues in their own writing. Additional scaffolds include L1 vocabulary support, prompt-simplification for struggling learners, and color-coded emoji feedback pointing to issues in accuracy, structure, flow, vocabulary, clarity, and evidence. The agent was iteratively pre-tested by two instructors to eliminate bypass loopholes before a two-week pilot following student preparation on critical thinking and ethical AI use. Grounded in observation, reflection, and feedback, results revealed initial learner confusion during demonstration as students expected conventional generative output, such as automated outlines. As students adapted to the constrained model, engagement increased especially among less-confident writers, who used the agent to identify their next steps without ceding decision-making ownership. The experience suggests that deliberately constraining AI to a reflective support role can shift learners’ interaction with generative AI from text production towards engagement with the writing process.
Keywords: ICT for Language Learning; Generative AI; L2 Writing; Learner Agency; AI Scaffolding
REFERENCES
[1] Selwyn, N. (2016). Education and technology: Key issues and debates. Bloomsbury Publishing.
[2] Warschauer, M., & Grimes, D. (2007). Automated writing evaluation: Encounters with machine feedback. Computer Assisted Language Learning, 20(1), 1–21.
