WorkBuddy Turns a Photo-Collage Idea Into a Live WeChat Mini-Program in Under an Hour, No Code Written
AI can now carry a non-programmer from idea to a published WeChat mini-program, but the real bottleneck is platform compliance: Tencent's 22-day filing dwarfs the hour-long build. For Western developers, the pattern confirms that regulatory review, not code generation, is what gates shipping inside WeChat's ecosystem.
WorkBuddy's AI agent walks a non-coder through the entire WeChat mini-program lifecycle from a single prompt. After rejecting an initial "slacking-off reading" idea as infeasible, the agent accepted a nine-grid photo collage tool, produced a feature list, asked clarifying questions, and generated a clickable HTML prototype. A subsequent "develop the mini-program from these two documents" prompt produced working code, with an optional cloud service toggle for data storage, login, uploads, or LLM calls.
The finished mini-program needed a missing "Creative Collage" module added via a follow-up prompt, and one screenshot-based fix request. WorkBuddy cannot simulate image selection, so final testing had to run in WeChat's official devtools. Registration cost 30 yuan for a personal account, then the app went through Tencent's review and a 22-day mandatory filing before going live. The guide insists on a separate workspace per project to keep the agent's memory clean and on completing the whole flow before filling knowledge gaps.
The 22-day filing wait versus the under-one-hour build shows that WeChat's compliance pipeline, not AI capability, is the true time sink for hobby developers.
The prompt pattern that tells the agent to ask about unclear points prevents scope drift and works because the human doesn't know what they haven't thought of.
AI agents still can't emulate touch interactions like picking a photo, so human verification through official devtools remains mandatory for UI-heavy mini-programs.
The AI-generated mini-program works, but its visual design is unfinished, as the author admits to using the first version without an AI design pass. The expectation is that a dedicated UI-focused skill should exist to iterate on the look, and a request for a concrete skill recommendation shows that such a workflow is missing. The snark in the opening question — why is AI output still ugly — captures the core complaint: code generation outpaces design polish.
It's all written by AI, why is the page still this ugly?
The UI wasn't adjusted by AI — it's still the first version.
Big shot, help me design one.
The UI should probably be redesigned; there should be a dedicated skill.
Recommend a skill; I want to learn it.