Generative Artificial Intelligence (AI) has revolutionized programming education but introduces significant risks of cognitive dependency

where students prioritize AI-generated code over algorithmic understanding. This study develops a task-centered guided interaction framework to intervene in this dependency. By integrating Design Thinking (DT) and Task-Centered System Design (TCSD)

we propose a transition from AI as a "result provider" to a "process facilitator". We developed a high-fidelity prototype tailored for Java courses

incorporating "beneficial friction" through step-by-step logic prompts and interactive checkpoints. Evaluation using the System Usability Scale (SUS) and Heuristic Evaluation with student participants (N=12) aims to validate the strategy's effectiveness in enhancing learning depth. Preliminary analysis suggests that guided interaction significantly reduces blind code duplication compared to traditional chat-based models.