Skip to content
Book Open access

Scaffolding Novice Programming in the LLM Era: Balancing Learning and Automation

Nov 2026 · 0 citations

Abstract

Large Language Models (LLMs) are reshaping programming, creating a dual challenge for computing education. While LLMs can help novices by translating natural language to code, there is a significant risk of over-reliance where learners get solutions without understanding the underlying concepts. This is compounded when AI tools "over-repair" student mistakes, hiding crucial learning opportunities. My PhD research addresses this challenge by designing and evaluating pedagogical frameworks that shift the focus from syntax-heavy coding to critical thinking and effective collaboration with AI. My work is guided by three research questions focusing on the design of LLM-powered educational tools, the analysis of student interaction patterns, and the evaluation of new pedagogical frameworks. My initial study evaluated NLPG-Exercise, a novel activity where novices arrange natural-language fragments to construct programs. This work established the importance of instruction fidelity—the principle that an educational tool must faithfully translate a learner’s logic, including mistakes, rather than silently correcting them. The study (N=170) demonstrated the feasibility of this approach but also revealed low but significant rates of fidelity errors, where the LLM either repaired incorrect logic or introduced new errors. Building on these findings, my future work will formalize these ideas into new pedagogical frameworks. This includes developing LLM Code Audit Problems, where students critique and refine flawed AI-generated code, and Agentic Prompt Problems, where they learn to solve complex tasks by collaborating with an LLM teammate. The ultimate goal is to develop and validate a cohesive pedagogical sequence that effectively teaches both programming and AI literacy, providing empirical evidence to guide curriculum design in the LLM era.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.