Building with AI: Top 10 Rules and Tools

Introduction

• Focus on building quickly with AI and reducing debugging time.
• Explore app building tools, frameworks, and MCPs.
• Course available with over 100 micro lessons.

Choosing the Right Tool

• Beginner to Advanced Tools:
  • Start with tools like Lovable and Bold New.
  • Progress to Replate Agent.
  • Advanced: Windsurf and Cursor.
• Pricing:
  • Mostly similar, but Lovable Bolt requires a subscription.
  • Free alternatives: Klein with VS Code (costs for model usage).
• General Setup:
  • File structure, code editing, terminal, and AI chat interface.

Planning with AI

• Avoid "vibe coding"; instead, create a structured plan.
• Resources:
  • Setup prompts at notes.switchdimension.com.
• Project Setup Steps:
  1. Talk to a Product Manager AI.
  2. Consult a UX Designer AI.
  3. Work with a Software Architect.
• Create and Review Documents:
  • Software Requirement Specification document.

Tools for Project Management

• Taskmaster AI:
  • Breaks down plans into subtasks.
• Use popular languages and frameworks for better AI support:
  • TypeScript, JavaScript, Python.
  • Frameworks: Nex.js, Tailwind CSS, Shad CN.

Step-by-Step Development

• Avoid one-shot prompts; prefer iterative development.
• Understand model strengths and weaknesses:
  • Use different models for code generation, planning, and troubleshooting.

Understanding Context Windows

• Be aware of model limits in context windows.
• Use document summarization tools in editors to manage context effectively.

Model Personalities

• Familiarize with model characteristics:
  • Example: Claude 3.5 vs. 3.7 for different tasks.
• Choose models based on task requirements (e.g., implementation vs. design).

Adding Context to Models

• Use tools like Windsurf and Cursor to add specific documentation context.
• Tools like Context 7 for accessing updated library docs.

Prompting Techniques

• Begin with a clear plan and step-by-step prompts.
• Use incremental development and review code changes regularly.
• Avoid complete reliance on "vibe coding."

Version Control and Testing

• Use git for version control over built-in checkpoint features.
• Recommended: Feature branching and committing changes.
• Begin writing tests with AI assistance:
  • Types: Unit tests, integration tests, and end-to-end tests.
• Use tests to ensure new features don't break existing functionality.

MCP Servers

• Start with essential MCP servers (e.g., Brave Search).
• Explore more as an advanced user.

Conclusion

• Transition from chaotic "vibe coding" to a structured, controlled process.
• Embrace the planning and tools to enhance app development with AI.