Agents that play, build, and test games.
49 use cases
Claude Code got an old DirectX game running on modern hardware, decompiled much of it, built a headless parallel runner, and now writes full level guides with screenshots.
A guide describes six bots building, shipping and improving a mobile game, including uploading ad creatives learned from a recorded click-path.
The poster built the browser 4X game Hexwardia in two weeks using Grok and Gemini to write the combat engine, fog-of-war, map generation and heuristic AI.
OnXStephen had Grok Bot spawn agents that built a Rust game engine with .NET scripting, multi-threading, weather, lighting, and hot reloading.
The poster built a game where Grok generates the story narration, and says it outperformed Gemini and GPT for narrative output.
Grok built the browser game Roofline, then trained a PPO agent that plays the live page, reaching a best score of 39,359.
The poster had Grok 4.5 build a level of an existing Three.js game from a detailed game design document, followed by some manual cleanup.
Four Grok 4.20 agents built a tower defense, mini Factorio, CHIP-8 emulator and roguelike shooter in parallel over a day.
Grok Build wrote the systems for a Mars simulator game, tested it, studied gameplay, fixed bugs and iterated.
Four Grok agents were asked to build an artificial life simulator with glowing particles, force sliders, and emergent behaviors.
A browser D&D 5e dungeon crawl where an AI dungeon master runs the game for a three-character party without inventing rules.
A commenter had Codex do fundamental architecture work for their game, reading engine docs and implementing better animation merging.
Two users had Replit Agent autonomously add 61 child-friendly, ad-free mini-games to their game vault, growing it from 40 to 101.
Agents including Claude Opus and o3 were given computers to independently play games like 2048 for hours.
Cowork handles design docs, task queues and playtests while Claude Code writes GDScript and runs Godot and git.
The author had GPT-5 in ChatGPT agent mode play online versions of Math Games with Bad Drawings, and it mostly worked, unlike earlier rental car and tax form attempts.
The OpenClaw team used computer-use agents to play Doom on remote machines, treating game performance as an informal way to improve the computer-use workflow.
Developers showed OpenClaw playing Forza Horizon on its own, imperfectly, and built playable bots for a multiplayer web game.