Generative AI for Developers: A Practical Guide to Prompting, Coding, and Collaboration

This chapter delves into the art of controlling generative AI through effective prompting. It distinguishes between system and user prompts, guiding readers on when to reuse prompts and how to create concise, one-off prompts. The importance of abstracting and refining reusable prompts is discussed. A key focus is on information strategy for prompts, breaking it down into three core elements for providing AI with appropriate context. Techniques like using bullet points for conditions, introducing constraints incrementally, and modifying prompts are explained. The chapter addresses how to handle AI that deviates from instructions and the power of role-setting to elicit expertise. It introduces few-shot and zero-shot prompting, and strategies for optimizing prompts based on context, balancing quality and length, and employing minimal prompts. The efficient use of English and Korean in prompts is also explored, along with using delimiters to separate context.

This chapter focuses on tailoring prompt strategies to different types of AI tools. For autocomplete-type AI tools, it emphasizes minimizing user prompts, supporting incremental implementation, and maintaining focus through rapid responses. Techniques like strengthening instructions with comments, providing and managing AI tool information, explicitly defining code, and pinning important files for immediate reference are discussed. For conversational AI tools, the focus is on flexible context control, supporting various file formats, accessing external information, and accumulating and reusing history. Clear prompting, early evaluation of prompt quality, AI-generated prompts, and AI-assisted refactoring are covered. The chapter also touches upon designing information for AI readability. For agent-type AI tools, strategies include pre-evaluating task suitability, adjusting granularity, partial delegation, and finding necessary tools.

This chapter explores development approaches designed to maximize the potential of AI. It discusses code architectures suitable for AI collaboration, such as reducing nesting for efficiency and designing AI-decoupled code. Scalability is considered in code design, along with applying systematic refactoring techniques and even re-implementing small open-source projects. Enhancing code quality with AI involves generating unit tests, defining clear test conditions, using decision tables for comprehensive test design, creating tests based on state transition diagrams, and removing redundant tests. The chapter also covers using AI for code reading, explaining logic with natural language and generating visual representations of complex logic. In code review, AI can be leveraged for performance improvements based on Big-O notation, code optimization using the BUD framework, evaluating data structure suitability, enhancing code quality based on SOLID principles, and employing Chain-of-Thought prompting.

This chapter provides practical tips for developers to effectively utilize AI in their daily work. It covers mastering editors and terminals, including removing unnecessary information from editors, utilizing automatic license checks, and leveraging integrated terminals. Tips for preventing AI hallucinations through help information and improving commit message quality using change diffs are also included. The section on data manipulation emphasizes AI-assisted regular expression generation, recognizing various date formats, generating POSIX CRON formats, converting special data formats, classifying unstructured data with AI, and streamlining data preprocessing. For rapid web development, AI techniques for SEO improvement suggestions and accessibility evaluation are presented. Finally, the chapter highlights essential tools for AI collaboration, such as using the diff command to identify changes, building and utilizing prompt libraries, converting to AI-friendly Markdown, creating readable diagrams with Mermaid, and making complex diagrams AI-readable with PlantUML.