Collaborative programming with Large Language Models (LLMs) like ChatGPT has growing potential for easier, faster, and more efficient coding and prototyping tasks, but users often encounter communication challenges that hinder their success. This study explores the psychological and interactional barriers in user-LLM communication, focusing on how users’ mental models and assumptions about AI capabilities shape their interactions. Patterns of miscommunication, such as ambiguous instructions, misidentification of issues, misaligned expectations, and emotional responses, prevent users from fully leveraging their problem-solving skills, and this project investigates how users could employ more effective problem-solving strategies with a clearer understanding of LLM limitations, leading to a more satisfying user experience.