Large language models (LLMs) hold great promise for enhancing teaching and learning in higher education, yet educators and administrators still lack practical examples to guide their adoption. This article presents insights and use cases from the integration of LLMs into a first-year undergraduate computer science cohort. By employing LLMs as digital scaffolds, timely support was provided helping students bridge knowledge gaps while engaging in independent problem-solving. At the same time, students were encouraged to maintain a critical stance by evaluating and verifying AI-generated content. These initial observations show that LLMs can encourage self-guided research, offer on-demand feedback, and strengthen cohort identity by acting as a mentor, peer, and liaison. Although the findings are exploratory, they serve as a point of reference for educators, informing future, more rigorous studies aimed at the successful integration of LLMs into higher education settings.