AI Coding Guide

Recommended Instructions for AI Coding Assistants

Learning how to effectively instruct AI coding assistants can significantly improve their output quality. This section provides best practices gathered from GitHub, OpenAI, and developer communities.

Table of Contents

Setting Context and Goals

Providing clear context and goals is essential for getting useful results from AI coding assistants:

  • Start general, then get specific: Begin with a broad description of the goal before listing specific requirements
  • Set the stage with high-level goals: Provide big picture context before diving into details, especially with blank files
  • Provide project context: Open relevant files and close irrelevant ones to help AI understand the codebase
  • Use clear file structure references: Explicitly mention file paths and relationships when working with multiple files
  • Specify programming language and frameworks: Clearly state which language, libraries, or frameworks you want to use

Writing Effective Prompts

The way you phrase your instructions significantly impacts the quality of AI-generated code:

  • Be specific and detailed: Clearly articulate desired context, outcome, length, format, style, etc.
  • Use separators for instructions: Use ### or """ to separate instructions from context
  • Break complex tasks into simpler ones: Divide large tasks into multiple simple, small tasks
  • Avoid ambiguity: Be precise about what you're referring to (specific functions, files, or code blocks)
  • Reduce "fluffy" descriptions: Replace vague descriptions with precise specifications
  • Say what to do, not just what not to do: Provide positive guidance rather than just restrictions

Tip

Instead of saying "Don't use deprecated methods," specify "Use the latest API methods from version X.Y, particularly the new methods for handling authentication."

Providing Examples and Structure

Examples and structure help AI models understand your expectations more clearly:

  • Give examples: Provide sample input data, expected outputs, and example implementations
  • Articulate desired output format: Show specific format requirements through examples
  • Use unit tests as specifications: Write tests first to define expected behavior
  • Use "leading words" for code generation: Start with import statements or language-specific keywords
  • Follow good coding practices: Use consistent style, descriptive names, and proper documentation

Iterative Improvement

Getting the best results often requires an iterative approach:

  • Start simple, then iterate: Begin with zero-shot prompts, then add few-shot examples if needed
  • Experiment and iterate: If you don't get desired results, refine your prompt and try again
  • Keep history relevant: Use separate threads for different tasks and delete irrelevant messages
  • Review and understand generated code: Take time to comprehend what the AI has produced
  • Test features thoroughly: Don't assume code works just because it looks good

Tip

When refining prompts, keep track of what works and what doesn't. Building a personal library of effective prompts can save time in future projects.

Technical Best Practices

Technical considerations can significantly improve AI code generation:

  • Use the latest models: Newer models tend to be easier to prompt and more capable
  • Consider temperature settings: Lower temperature (0) for factual tasks, higher for creative tasks
  • Manage context window efficiently: Be mindful of token limits and prioritize important information
  • Use appropriate tools for different tasks: Choose the right AI tool based on the specific coding need
  • Break down code into smaller functions: AI performs better with smaller, modular code components

Security and Quality Considerations

Ensuring security and quality of AI-generated code requires vigilance:

  • Always review generated code: Check for security vulnerabilities and logical errors
  • Verify external API references: Confirm that referenced methods and libraries actually exist
  • Test edge cases: Ensure code handles unexpected inputs and boundary conditions
  • Check for outdated patterns: Verify that generated code follows current best practices
  • Maintain transparency: Be clear about what parts of your code were AI-generated

Tip

Consider implementing automated security scanning tools in your workflow to catch potential vulnerabilities in AI-generated code.