AI Programming / Vibe Coding (Pair Programming)

Vibe Coding is a way of collaborating with AI during programming — the AI acts as a pair programming partner, helping developers complete development tasks more efficiently. This approach changes the traditional programming workflow, letting developers focus more on design thinking and business logic while handing off tedious implementation details to the AI.

AI Models

AI Coding Capability Blind Test
GLM (Zhipu): Currently my daily-driver model — I’ve purchased their coding plan. Very affordable, and I also share it with the Lobster Bot.

AI Coding Tools I’ve Tested and Used (Game Client Perspective)

Cursor

• Highlights: A standalone AI coding IDE (built on VS Code)
• Strengths: Supports multiple models, friendly UI, powerful code completion
• Best for: Full-stack development, rapidly scaffolding new projects

Claude Code

• Highlights: CLI tool + VSCode plugin
• Strengths:
  • Strong knowledge of third-party models
  • I’ve already configured it with Zhipu GLM
• Best for: Projects requiring flexible configuration and command-line operations
• Game Dev: Well-suited for server-side development and toolchain work

GitHub Copilot

• Highlights: Integrated into IDEs (VS, Rider)
• Strengths:
  • Most game-developer-friendly (supports Rider)
  • Real-time code completion
  • Multiple model options available
• Plan: Using Copilot Pro
• Best for: Day-to-day development, code completion
• Limitation: So-so compatibility with third-party models

The TFCDC Methodology

TFCDC is the core methodology for Vibe Coding, helping developers collaborate effectively with AI.

Think

Think thoroughly and plan before you start coding.

Core Questions

1. Project Essence

   • What is this project?
   • What is the core value?
   • Who are the target users?

2. Gamified Thinking

   • Treat programming as a “game”
   • Define the rules and objectives of the game
   • Understand the win conditions

3. Execution Logic

   • What is the gameplay loop? (How do you execute?)
   • What is the play strategy (tactics)?
   • What is the critical path?

Thinking Template

1
2
3
4

Project name: ___________
Core objective: ___________
Technical challenges: ___________
Priority: ___________

Example Diagram

Framework

Define your technology choices and architectural design.

Tech Stack Selection

• Frontend: React / Vue / Unity / Unreal
• Backend: Node.js / Go / Python
• Database: MySQL / MongoDB / Redis
• AI Integration: OpenAI / Claude / local models

Architectural Design

• MVC Pattern: Model-View-Controller
• Microservices: Service decomposition and communication
• Event-Driven: Event-based architecture
• Component-Based: Reusable component design

Decision Process

1. First, understand the project requirements
2. Evaluate technology options
3. Weigh the pros and cons
4. Make a decision
5. Communicate your decision to the AI

💡 Tip: If you’re unsure about tech stack choices, ask the AI for its recommendation first

Checkpoints

Use version control to establish key checkpoints.

Version Control Best Practices

• ✅ Must use Git
• ✅ Commit after each feature is complete
• ✅ Write clear commit messages
• ✅ Use branches for feature development
• ✅ Create milestone tags regularly

Commit Message Convention

1
2
3
4
5
6

feat: add new feature
fix: fix bug
refactor: code refactoring
docs: documentation update
test: test-related
chore: build/toolchain-related

Checkpoint Examples

• 🚀 M1: Base framework complete
• 🚀 M2: Core features implemented
• 🚀 M3: UI/UX complete
• 🚀 M4: Tests passing
• 🚀 M5: Release ready

Debugging

Effective debugging strategies and problem-solving approaches.

Debugging Process

1. Find the problem yourself

   • Reproduce the bug
   • Collect error information
   • Locate the problematic code

2. Let the AI propose solutions

   • Provide the full error context
   • Describe expected behavior
   • Supply relevant code snippets

3. Validate the solution

   • Understand the AI’s suggestion
   • Apply changes carefully
   • Test and verify the result

Understand the Project Structure

• 📁 Understand the directory structure
• 📁 Familiarize yourself with key files
• 📁 Know the dependency relationships
• 📁 Master the configuration files

Debugging Tips

• Use logging
• Add breakpoint debugging
• Verify with unit tests
• Use bisection to isolate the problem

Context

Providing the AI with sufficient context is the key to getting high-quality code.

Types of Context

PRD (Product Requirements Document)

• Definition: A document describing product features, user experience, and business requirements
• Contents:
  • Functional requirements
  • User stories
  • Acceptance criteria
  • Priorities
• Example:

  1 2 3 4

  ## User Login Feature - Support login via email / phone number - Verification code authentication - Keep me logged in (7 days)

MVP (Minimum Viable Product)

• Definition: The version with the minimum features needed to validate a product hypothesis
• Principles:
  • Minimize features
  • Iterate quickly
  • Validate assumptions
• Example:
  • ✅ Included: core features
  • ✅ Included: basic UI
  • ❌ Not included: advanced features
  • ❌ Not included: full optimization

Technical Documentation

• API documentation
• Database design
• System architecture diagrams
• Code comments

Project Background

• Project goals
• Technical constraints
• Time limits
• Team situation

How to Provide Effective Context

✅ Good Context

1
2
3

This is a Unity game project developed in C#.
I need to implement an enemy AI that chases the player when they enter within 5 meters.
The enemy already has a NavMeshAgent component and needs to use NavMesh pathfinding.

❌ Bad Context

1

Write me an AI.

Tips for Collaborating with AI

1. Clear Instructions

• ✅ “Implement a to-do list with create, read, update, and delete support”
• ❌ “Make something”

2. Work Step by Step

• Break large tasks into small steps
• Implement incrementally and verify as you go

3. Code Review

• Don’t blindly accept AI-generated code
• Understand what every line does
• Ask the AI to explain when necessary

4. Iterative Improvement

• Implement the basic functionality first
• Then optimize and refine gradually

Project Management

Task Planning

1. List all to-do items
2. Evaluate priorities
3. Estimate workload
4. Set a timeline

Progress Tracking

• Use TODO lists
• Review progress regularly
• Adjust the plan in a timely manner

Concept Explanations

Orchestration (Swarm Mode)

Complex orchestration + master controller + swarm.
Whoever masters orchestration can command the swarm.

Explanation: In AI programming, Orchestration refers to coordinating and managing the work of multiple AI models or tools through a master controller. Just like a beehive where the “queen” issues commands and multiple “worker bees” divide up tasks to collaborate — this approach enables:

• Different models handling different roles (e.g., one for code generation, one for debugging, one for testing)
• Automated workflow orchestration
• Concurrent processing of multiple sub-tasks
• Unified result aggregation

MCP Framework

Explanation: MCP (Model Context Protocol) is an open protocol standard for communication between AI models and external tools and data sources. It defines a unified interface that allows AI to access external resources safely and in a standardized way. Core values include:

• Standardized tool call interfaces
• Secure permission control mechanisms
• Cross-platform compatibility
• Plugin-style extensibility

Token

The basic unit by which AI models process text. Roughly equivalent to 3/4 of an English word or 1 Chinese character. The token limit of the context window directly affects how much information the AI can understand.

Token Calculation Reference

• Chinese: ~1 character = 1 token
• English: ~4 characters = 1 token
• Code: ~4–6 characters = 1 token (varies with code complexity)
• Newlines/spaces: Usually free or cost very few tokens

Estimation Methods

1. Rough estimate: Character count ÷ 4 ≈ token count
2. Accurate calculation: Use the token counting tool provided by the model
3. Context consumption:
   • Input prompt: calculated by the rules above
   • Output content: typically 50%–200% of input tokens
   • Conversation history: accumulates into the context

Practical Tips

• Replace long code files with summaries
• Remove unused comments and blank lines
• Periodically clear conversation history
• Keep an eye on context window limits per model (e.g., 4K / 8K / 32K / 128K)

Common Pitfalls

❌ Avoid These Mistakes

1. Over-reliance on AI

• Don’t give up thinking entirely
• Maintain your understanding of the code
• Make key decisions yourself

2. Insufficient Context

• Not providing enough information
• Making the AI guess your requirements
• Getting unsatisfactory results

3. Blindly Copying Code

• Using code you don’t understand
• Introducing security risks
• Hard to maintain

4. Ignoring Version Control

• Not committing code promptly
• Missing backups
• No way to roll back

Advanced Tips

Improving AI Coding Efficiency

1. Build Code Templates

• Commonly used code snippets
• Project structure templates
• Configuration file templates

2. Create a Prompt Library

• Prompts tailored to different scenarios
• Verified, effective instructions
• Reusable context templates

3. Use Tool Integrations

• Git hooks
• CI/CD integration
• Automated testing

4. Team Collaboration

• Share prompts
• Unify code style
• Build a knowledge base

🎯 Recommendation Guide

📚 Learning Resources

• MCP Official Docs: https://modelcontextprotocol.io/
• MCP Registry: https://registry.modelcontextprotocol.io/
• Video Tutorials: The GenOrca project includes a YouTube demo video

🔍 Search Keywords

• unreal mcp
• unreal engine mcp server
• ue5 model context protocol
• unreal ai automation

Learning Log

September 2025 and earlier

• Studied the Vibe Coding methodology
• Used Tencent’s internal coding assistant along with Copilot and Cursor
• Researched the TFCDC framework

January 2025

• Tried out Claude Code and started using MCP

February 2026

• Researched and compiled notes on the UnrealMCP project

To Do

• Flesh out AI use cases for game development scenarios
• Organize more code templates
• Collect commonly used prompts