AI-Assisted Help-Seeking Trajectories in Programming Education from an SRL-Informed Perspective

A study of 71 university students' interactions with generative AI in introductory Python programming reveals that most use AI reactively for troubleshooting rather than as a planned learning tool. While AI-assisted help-seeking patterns didn't significantly affect task scores, they substantially influenced the number of code submissions required, suggesting that how students engage with AI matters more than whether they use it.

This research examines a critical tension in modern education: generative AI's capacity to accelerate learning versus its potential to undermine self-regulated problem-solving skills. The study analyzed 1,290 student prompts and 17,190 code submissions using a self-regulated learning (SRL) framework, finding that students predominantly employ AI for reactive debugging rather than intentional, structured problem-solving approaches. This distinction carries significant implications for educational outcomes beyond simple performance metrics.

The findings emerge as universities worldwide grapple with AI integration in computer science curricula. Previous scholarship focused narrowly on code correctness and overall AI usage frequency, missing the nuanced question of how students structure their help-seeking behaviors over time. By tracking trajectory patterns—how students progress through multiple interactions and attempts—the research reveals that students with more efficient engagement patterns require fewer code submissions, indicating that learning efficiency depends on help-seeking strategy rather than raw performance.

For educators and edtech developers, these results suggest that AI tool effectiveness hinges on supporting intentional learning workflows. Students who ask conceptual questions upfront and use AI for planned exploration likely develop stronger problem-solving capabilities than those jumping between reactive debugging cycles. This has direct implications for how educational platforms design AI interfaces: prompting students toward reflective, structured help-seeking could enhance learning outcomes without restricting access.

Looking forward, institutions implementing AI-assisted programming education should monitor not just student grades but help-seeking patterns and submission frequency as proxy indicators of learning quality. Curriculum design might incorporate explicit instruction on productive AI engagement, treating help-seeking strategy as a metacognitive skill.

• →Students primarily use generative AI for reactive troubleshooting rather than planned, self-regulated problem-solving in programming tasks.
• →Help-seeking trajectory patterns correlate with submission efficiency but not with task scores, suggesting learning quality varies by engagement approach.
• →The number of code submissions required indicates help-seeking effectiveness, with efficient trajectories requiring fewer iterations.
• →Educational impact of AI depends on how students structure help-seeking interactions across multiple turns and attempts.
• →Current AI tool design may inadvertently encourage reactive debugging cycles rather than deeper conceptual problem-solving.