Letting AI play my game – building an agentic test harness to help play-testing
blog.jeffschomay.com
April 29, 2026
8 min read
🔥🔥🔥🔥🔥
53/100
Summary
AI is being utilized to play-test the game Crossword Dungeon alongside the developer, improving the validation of the game's functionality and bug detection. This method enhances the efficiency of the play-testing process compared to traditional manual testing.
Key Takeaways
- The author developed a method for AI to play-test their game, Crossword Dungeon, which significantly improved the efficiency of the testing process.
- The AI was able to create a node.js wrapper that allowed it to interact with the game without modifying the original files, using a text-based renderer for navigation.
- The AI demonstrated the ability to identify bugs and recreate specific scenarios, effectively mimicking human player behavior and strategic insights.
- The game combines crossword puzzle mechanics with RPG elements, where solving letters levels up connected rooms, increasing the game's difficulty.
Community Sentiment
Positives
- Using AI agents for balance testing in games can significantly streamline the process for indie developers, allowing for more efficient play-testing and game refinement.
- The ability to automate interactions and fix regressions while developing games with AI tools like Copilot CLI enhances productivity and reduces manual testing efforts.
- Creating a CLI for AI to play games allows for innovative end-to-end testing strategies, enabling developers to assess game functionality in real-time.
- Having AI verify its own work through end-to-end tests can lead to more reliable outcomes and allows developers to progress even while they are away.
