Vibe Coding a FULL Game 🎮 (AI Coding)

Summary

Matthew Berman demonstrates building a complete 2D, grid-based, turn-based sci-fi strategy game through “vibe coding,” without manually writing code. He used Grok to develop the game concept and rules, then Cursor with Claude 3.7 Thinking to create and iteratively improve the game.

The project, named Nebula Dominion, took about four hours, much of which was spent waiting for the AI to make changes.

Game features

• A 10×10 grid with four unit types: Corvettes for scouting, Mechs for ground combat, Dreadnoughts for heavy attacks, and Drones for gathering resources.
• Resource management using “nebula,” including unit-building costs and a one-turn wait before newly built units can be used.
• Turn-based movement, warp lanes, combat between adjacent enemy units, and attackable bases. Destroying the opposing base is the main win condition.
• An AI opponent whose moves are slowed down so the player can follow them.
• Tooltips showing unit stats and movement ranges, along with interface improvements such as automatically switching the selected unit.
• Sprite-based graphics for units, planets, and map tiles. Berman downloaded an asset folder and asked Cursor to match files to game objects.
• Debug controls for quickly testing win conditions, followed by AI-assisted refactoring.

Guidance and workflow

Berman’s main advice is to write a clear specification before building. Using inconsistent terms—such as “base” and “command center” interchangeably—can lead to confusion in the code and require cleanup later.

His recommended development cycle is:

1. Make a change and test it manually.
2. Ask the AI to write appropriate tests, especially integration tests.
3. Confirm the tests pass.
4. Commit and push the working version before moving on to the next feature.

He emphasizes committing frequently because changes can unexpectedly break other parts of a project, and Git commits provide useful recovery points. He also describes the process as asynchronous: he could leave Cursor working for several minutes, sometimes up to ten, while doing other tasks.

Sponsor segment

The video includes a promotion for LangTrace, described as an open-source, OpenTelemetry-based observability and evaluation platform for AI applications. Its features include collecting traces and datasets, running evaluations, and monitoring LLM, agent, framework, and vector database activity. The segment also mentions a hosted version, a discount, and webinars.

Main speakers and tools

• Matthew Berman — presenter and creator of the game.
• Grok — used to generate the initial game concept and rules.
• Cursor with Claude 3.7 Thinking — used to build, debug, refine, and refactor the game.
• LangTrace — sponsor featured in the video.