This isn't a "developers use AI to be more efficient" story. It's a restructuring of how work gets done: a product manager using AI to do what once required a whole team.
She is Qi Yi. Over the past six months she completed three full app projects with AI, and within two months migrated her complex community mini program into an app. Behind this isn't just tools — it's an "AI collaboration method" distilled from repeated trial and error.
Five years of operating, breakthrough via AI
She runs a vertical community mini program with very complete features — community posting, message notifications, a membership system, shop payments, inventory items… almost every community-ecosystem feature you can imagine.
The problem was real: users came, but every next step — promotion, operations, iteration — required sustained investment. At the startup stage money was tight, no team could be supported, and community growth stalled. It was nearly a dead end.
AI's rapid progress showed her a way out — she could fully support all the work with herself plus N AIs.
The hardest start: first convert the mini program into an app, then use AI for all operations and promotion. The first question was — how do you build an app with AI?
Her first step: let AI "read" first
Many people's first instinct is "tell AI what I want to do." Qi Yi's approach is the opposite — first let AI understand her existing product.
She uploaded the mini program's source code to GitHub, then had AI (Codex/Cursor) reverse-produce the product feature docs, sorting out "what I already have" before continuing.
Two benefits: it avoids the tediousness of manually re-describing features, and it ensures AI's understanding of the product is based on actual code rather than vague natural language.
Safety first: build walls before inviting AI to work
This is the most instructive part of Qi Yi's method. Before any development, she first did environment isolation with Tencent Cloud CloudBase.
This seemingly simple "build walls first" strategy is exactly what many AI developers overlook. It's not that AI shouldn't work — it's about drawing safe boundaries for it.
Zero technical experience, driving AI to meet needs accurately
Through the whole development process, Qi Yi also has a unique AI communication strategy: first discuss technical approaches with ChatGPT, distill professional terms, then pass them to the executing AI as keywords. ChatGPT is the architect, Cursor is the engineer — never let them do the same job, keeping context clean.
Compare and you'll see:
- ❌ "Make my list load a bit faster, you know, prepare the content before the user scrolls"
- ✅ "Implement lazy loading and preloading optimization"
Quality verification: three lines of defense against AI mistakes
Can AI-written code be trusted? When a bug survived several rounds of debugging, Qi Yi's move was: reverse-description verification, log debugging, and architecture convergence.
The coolest step: managing operations through conversation
Qi Yi didn't want to open a computer and log into a CMS every time she changed an operations config. She wanted: one sentence to complete operations management.
This might be the most "non-technical thinking" innovation in the whole project. She described the workflow she wanted to AI: receive natural-language operations instructions, generate a config draft, write the config to the database directly through Tencent Cloud CloudBase MCP, and return the result.
AI wrote a Skill accordingly. Now she types one sentence in the dialog and the banner changes.
Tencent Cloud CloudBase MCP is the key here: it lets AI operate cloud resources directly — insert or modify config data — without writing API integration code, without manually configuring permissions. Operations through natural-language conversation.
Real review: Tencent Cloud CloudBase vs SupaBase
As a non-technical user who has used multiple platforms, her review is direct:
| Dimension | SupaBase | CloudBase |
|---|---|---|
| Interface friendliness | More technical jargon | Fewer technical terms, easier to understand (recommended) |
| Onboarding barrier | Needs some technical background | Fits non-technical users |
| AI integration | Do it yourself | MCP + Skills out of the box |
| WeChat ecosystem | Not supported | Deep mini program / official account integration |
Three wishes for Tencent Cloud CloudBase
As a deep user, she also offered candid expectations: out-of-the-box operations Skill templates, AI ops and alert notifications wired together, and automatic analysis with auto-generated daily reports.
Her AI development toolchain
The whole project runs on a carefully assembled toolchain — a combination of LLM + skill orchestration + a self-developed store-management app.
She's also promoting another app she developed: an app-management assistant based on AI + Tencent Cloud CloudBase MCP that generates cloud functions to interface with Apple's API, solving app management and global adaptation for the Apple App Store.
In closing
Qi Yi's practice reveals a broader trend: AI programming is lowering the barrier to development, letting people "with ideas but no coding" build products too.
Her methodology wasn't learned from a textbook — it was distilled step by step through real projects.
Her expectations also show us something: when AI doesn't just write code for you but also handles ops, operations and decision-making, "non-technical people building products" will no longer be an exception.
