LLM coding assistants have transformed software development, but using them effectively demands a clear strategy. Developers must approach these tools like skilled pair programmers, providing context and direction. The author emphasizes the importance of starting with a detailed specification before writing any code. By brainstorming and outlining requirements with the AI, developers can create a comprehensive spec document that guides the coding process. This upfront investment, though it may feel slow, significantly streamlines subsequent coding efforts. Breaking projects into small, manageable tasks is another key practice. The author notes that large, monolithic requests often lead to confusion and messy outputs. Instead, handling one function or feature at a time allows the AI to produce clearer, more coherent code. This iterative approach not only aligns with good software engineering practices but also enhances the AI's performance. Several coding-agent tools now support this chunked workflow, enabling developers to create structured prompts for each task. Providing ample context is essential for optimal AI performance. Developers should supply all relevant information, such as existing code, project constraints, and specific goals. Tools like Claude can integrate entire GitHub repositories into their context, but the author recommends going further by manually including key documentation or code snippets. This ensures the AI operates with complete information, reducing the risk of errors. Emerging utilities that automate context packaging can also help manage larger projects, allowing developers to bundle essential files for the AI to process. The goal is to ensure the AI has all the necessary facts at its disposal, leading to more accurate and effective coding outcomes.