Two AI-built HTML game demos in 210 minutes
The author used free ChatGPT to iterate three versions of development documents for Flycard, then handed the refined document and initial buggy code to Tencent Yuanbao; the game reached four versions. They next used Tencent Yuanbao to build Impossible Dragon, rejecting two proposed designs and fixing two bugs over five rounds, with each personally played round lasting under 50 seconds. Both demos were uploaded to a code-hosting platform rather than formally released; the author reports 210 minutes in total, including under 90 minutes of active work on Flycard and under 120 minutes on Impossible Dragon.
This first-person case study offers concrete evidence that current models can help newcomers build and iterate small HTML game prototypes, while also identifying the need for human feedback and testing as project scope and production complexity increase.

注:文章本身没有AI辅助,游戏确实是AI做的。
不论你是否有才华,或者是否有想做游戏的冲动,齐备的物质条件都在说明一件事:今天,任何人都可以很快地做出一款,甚至多款电子游戏。
在笔者的传统认知里,想要做出一款游戏,怎么说都要花上几天构建关卡,还要花心思打磨,统筹美术资源和音乐,怎么说好像都不是完全没接触过游戏开发的人能上手的东西。
但很明显,笔者弄错了做游戏的方法。
产品研发中有一个术语叫MVP(Minimum Viable Product),翻译成中文是“最小可行性产品”,不是那个什么NBA球场上的MVP(Most Valuable People,最有价值球员)。在绝大多数情况下,很多游戏就是从一个堪堪能玩的小Demo逐步开发成成品的。尽管游戏非常关注“打磨”,但这个“从零到一”的启动步骤也是至关重要的。