跪拜 Guibai
← All articles
Cocos Creator · Game Development · AI Programming

Shipping a WeChat Mini Game With Zero Gamedev Experience, Using AI as a Full Team

By 技术更好说 ·
Read original on juejin.cn ↗ Google Translate ↗ Alt translation

The post maps out a concrete, end-to-end workflow where a single person substitutes AI for an entire product team, then hits the real ceiling: AI can build a correct system, but it cannot judge fun, difficulty pacing, or whether a 6-year-old can actually use the interface. That judgment gap is where solo makers and small teams will win or lose.

Summary

The game, "English Word Storage Box," has 150 levels and over 800 words, built in Cocos Creator. AI was not just a code generator but acted as product manager, game designer, UI/UX assistant, programmer, and tester. The maker structured word banks and rules, then let AI review level configurations, generate prototypes, write game logic, and run automated test scripts.

The real friction surfaced after the code worked. Program correctness and game feel turned out to be separate evaluation systems. AI could batch-generate 150 valid levels, but only repeated human playtesting revealed that the difficulty curve was flat and boring. Similarly, a UI that made functional sense for a developer failed for a 6-year-old who might not yet read fluently.

Launching on WeChat's mini game platform added another layer of work: platform adaptation, package checks, real-device testing, review materials, and handling rejections. The gap between a working demo and a shipped product was far wider than expected. The next phase focuses on instrumentation and real user data to drive difficulty tuning, feedback polish, and retention experiments.

Takeaways
AI handled five distinct roles on this project: product manager, game designer, UI/UX assistant, programmer, and tester.
Structuring the word bank and classification rules let AI audit 150 levels for difficulty jumps, category conflicts, and child-appropriateness, turning manual drudgery into a rules-plus-data pipeline.
Program correctness and game experience are separate evaluation systems; AI-generated levels were logically valid but boring, a problem only human playtesting could catch.
A UI that works functionally can still fail a 6-year-old who cannot read well, a design blind spot that emerged only when a friend tested with real kids.
Moving from a local Cocos Creator build to a live WeChat Mini Game required platform adaptation, package size checks, real-device testing, review submissions, and handling rejections.
The next development phase will instrument the game with data and feedback loops before adding features, letting real player behavior drive difficulty curves and retention experiments.
Conclusions

The most instructive part of this build is not that AI wrote the code, but that it absorbed the coordination overhead usually spread across five people, letting a solo maker operate at team scale.

AI's inability to judge fun or age-appropriate UX is not a temporary limitation; it is a hard boundary. The maker's taste and willingness to watch a child struggle with the interface become the product's only quality control.

The post's distinction between 'making learning seem fun' and 'making a game fun so learning happens inside it' is a design principle that applies far beyond educational games, to any tool that must compete with distraction.

Shipping to a gated platform like WeChat Mini Games adds a compliance and operations tax that AI cannot pay; the solo maker still has to walk through every review step personally.

Concepts & terms
Cocos Creator
A cross-platform game engine popular in China for 2D and lightweight 3D games, with first-class support for WeChat Mini Games.
WeChat Mini Game
A lightweight game format that runs inside the WeChat app without a separate install, subject to platform review, package size limits, and WeChat-specific APIs.
Source: juejin.cn ↗ Google Translate ↗ Backup ↗