Introduction
This research explores the emerging practice of "vibe coding"—the creation of software using natural language rather than traditional programming code—within the context of English as a Foreign Language (EFL) education. As Large Language Models (LLMs) become ubiquitous, the ability to "metalanguage" with AI—bridging the gap between creative intent and functional output—becomes a critical literacy for students.
The Challenge: From Consumers to Creators
While students frequently use AI to generate text, using it to design functional tools presents a higher cognitive hurdle. The authors identify a significant gap in current educational approaches: interaction with AI is often treated as a simple query-response cycle. However, "vibe coding" requires students to act as designers, necessitating a sophisticated understanding of how to prompt, negotiate authorship, and conceptualize the technology itself.
Methodology
The researchers conducted a qualitative pilot study involving a four-hour workshop for two high-achieving secondary school students in Hong Kong. Following a "Backward Design" approach, students were tasked with building an app to assist peers with the HKDSE English writing exam.
The study employed a Human-AI Metalanguaging Framework to analyze the interaction across three dimensions:
- Talking to AI: The specific prompt engineering strategies used.
- Talking through AI: How students negotiated their own voice and authorship.
- Talking about AI: The internal mental models students held regarding the nature of AI.
Data was collected through think-aloud protocols, screen recordings, and a "draw-a-picture" technique where students generated AI images to represent their conceptualization of the technology.
Key Insights: The Architect vs. The Collaborator
The study revealed that a student's mental model of AI fundamentally dictates their success in building functional applications.
- Prompting Styles: Student A (the "Architect") viewed AI as a precise tool, preparing long, structured prompts offline (averaging 554 words). In contrast, Student G (the "Collaborator") treated AI as a conversational partner, using short, iterative prompts (averaging 21 words).
- The Authorship Paradox: A significant finding was the difficulty students faced in accurately perceiving their own authorship. Despite intending to provide the majority of the ideas, the high cognitive load of vibe coding often blurred the lines between the student's voice and the AI's output.
- AI as a Languaging Machine: Vibe coding serves as a powerful reflective tool. It forces students to regulate and articulate their logic and language precisely to ensure the machine produces the desired outcome.
Results
The divergent approaches led to markedly different outcomes:
- Student A: Successfully developed "Your Friendly Learning Buddy," a functional application that provided specific scores and qualitative feedback based on official exam guidelines. Her systematic prompting ensured the final product aligned with her original vision.
- Student G: Produced "Writing Quest," which remained a static webpage. Her conversational approach led to technical errors and "paywall" barriers in the freemium tools used, highlighting how prompting strategies impact resource management.
Conclusion and Future Directions
The study concludes that vibe coding instruction must be explicit. Educators should move beyond teaching basic prompting to help students develop a "critical vocabulary" for AI. Future work should focus on facilitating deeper discussions regarding authorship and the rhetorical load shared between humans and machines, perhaps starting with simpler, non-multimodal projects to build foundational skills.
Summary updated