Large language models (LLMs) have transformed the way people write code. Yet current interactions with LLMs in programming largely remain linear, where programmers issue prompts, the LLM passively produces responses, and programmers continue with follow-up prompts in a back-and-forth cycle. Inspired by the grounding theory in communication where humans iteratively establish a mutual understanding via repetition, clarification, and emphasis, we explore how similar principles can enrich human-LLM interaction. In this talk, I will present methods that enable the LLM to explain its generated code and ask clarification question and also enable humans to emphasize important instructions via attention steering. I will also demonstrate how these methods improve programming productivity, foster trust, and enhance the overall programming experience across multiple settings.