While Large Language Models (LLMs) enable experienced programmers to increase their productivity, LLMs’ impact on learning and productivity for novices is currently unclear. Recent work showed novice programmers struggle with prompting LLMs for code generation and suggested that the use of LLMs in CS education could exacerbate existing equity issues. Educators are now faced with the difficult question of whether and when to incorporate the use of LLMs into the CS curriculum without adversely impacting student learning and equity. To address these concerns, we study the effects of using an interactive LLM on code generation with novice programmers. We find that using our interactive LLM improves the accuracy of code generation over the baseline LLM. Additionally, after using the interactive LLM, novices write improved prompts even when using the baseline LLM. Based on our findings, we plan to create iGPTs, a set of customized, interactive LLMs spanning computational learning goals as templates to facilitate LLM integration for improving student learning and retention.