How I Built a Full Xiangqi Game in a Week Using GitHub Copilot (Instead of 3 Months with a small Team)

Source: Dev.to

Background

Three years ago, building a Xiangqi (Chinese Chess) game required a team and months of development. Today, I rebuilt it almost entirely by myself—with the help of GitHub Copilot.

You can try the game here: Cờ Tướng

This article isn’t about hype; it’s about the real workflow changes, practical lessons, and how AI fundamentally changed how I build software.

Previous Development Process

• Team composition: 2 developers, 1 designer
• Timeline: ~3 months

Front‑end work

• Login page, register page, lobby, gameplay screen, table‑management system
• Designer responsibilities: board textures, piece designs, UI layout, lobby visuals, buttons, panels, effects
• Developers focused on:
  1. UI implementation – building pixel‑perfect UI from design files
  2. Multiplayer system – room creation, joining tables, synchronizing moves, handling reconnects
  3. AI engine – move generation, position evaluation, difficulty levels, avoiding obvious blunders
  4. State management & bug fixing – illegal moves, disconnections, game‑ending logic, synchronization errors

Even releasing a basic playable version took around three months.

Rebuilding with GitHub Copilot

Team shift

• No designer
• Only me and Copilot

My role changed from writing code to directing code.

New workflow

1. Think of an idea
2. Write a prompt
3. Review Copilot’s output
4. Refine with additional prompts
5. Repeat

Treat Copilot like a junior developer who is extremely fast but needs clear direction. It’s not magic, but it’s incredibly productive.

Prompting Strategies

Bad prompt

Create a full lobby UI for my game

Result: messy, generic, unusable.

Good prompt

Create a TableManagementPage with:
- wooden texture background
- table list with pagination
- each table shows name, time limit, turn limit
- user can click to join a table
Use React and CSS modules

Result: high‑quality, ready‑to‑use component.

Because the scope is clear, Copilot delivers focused code.

Backend example

Prompt

Create API endpoints for TableManagementPage

Copilot generates

• GET /api/tables
• POST /api/tables
• JOIN /api/tables/:id
• validation logic
• data models

Since it already understands the existing codebase, you don’t need to describe every detail; context is enough.

UI example

Prompt

Create a wooden‑texture Xiangqi board UI with traditional Chinese style

Follow‑up prompt

Make the board darker

Copilot produces a usable UI instantly; I only tweak small details.

What Copilot Excels At

• Writing React components
• Generating API endpoints
• State management and validation logic
• Repetitive UI code

These tasks, which previously took hours, now take minutes.

What Copilot Is Not Good At

• Complex game algorithms
• High‑level architecture decisions
• Long‑term consistency across the codebase
• Optimizing AI engine logic

You still need solid engineering experience; Copilot merely accelerates development.

Workflow Comparison

Results

• Previous: 2 developers + 1 designer → ~3 months
• Now: 1 developer + GitHub Copilot → dramatically less time

The shift is real, not theoretical. Copilot removes most repetitive work, letting me focus on architecture, game logic, and final decisions.

Conclusion

The future of development isn’t AI replacing developers; it’s developers using AI to build faster than ever. If you’re not incorporating AI into your workflow, you’re already behind.

Feel free to comment with questions or to discuss building games with AI.

You can try the game here: https://cotuong.club/