Basics
| Item | Details |
|---|---|
| Founder | Chen Yunfei (economics background, former internet-company department head, Huazhong University of Science and Technology) |
| Product | Cat Light (小猫补光灯) — an AI photography assistant app |
| Team size | 1 person |
| Revenue | ~¥1M/year; Pro version hit #1 paid photography app on the App Store within 24 hours; 100K+ downloads |
| AI stack | Claude 3.5 + Cursor (built in 1 hour) |
| Source | 36kr / Woshipm |
Storyline
Background: Chen Yunfei is a former internet-company department head with no coding background (economics degree). He had always been interested in building indie products, but the technical barrier kept blocking him. AI coding tools opened the door.
Turning point: Chen used Claude 3.5 + Cursor to build the Cat Light app in 1 hour. This wasn’t a toy demo — the paid Pro version hit #1 on the App Store paid photography chart within 24 hours of launch and racked up 100K+ downloads. The key was his product instinct: he didn’t build something “technically cool but unused.” He nailed a real need from photography enthusiasts on Xiaohongshu — selfie fill light on the phone. Product instinct + AI coding = a non-technical person can ship a hit app.
Outcome: ~¥1M annual revenue. He spends only half his day on the product and travels the rest of the time. Revenue is diversified — App sales + Xiaohongshu personal media + consulting + teaching. It is the benchmark case in China for non-technical founders building products with AI.
Workflow Breakdown
Minimum loop analysis
| Step | What he did | AI involvement | Notes |
|---|---|---|---|
| 1. Think it through | Found the “selfie fill light” pain point from Xiaohongshu users | 0% | Product instinct came purely from observing the user community |
| 2. Write a spec | Described app features in natural language | High | A non-technical spec is “I want an app that does X” |
| 3. Implement | Claude 3.5 + Cursor produced a complete app in 1 hour | Very high (~95%) | Zero to shippable in 1 hour |
| 4. Validate | Listed on App Store + Xiaohongshu promotion | Low | Hitting #1 in 24 hours is the most direct validation |
| 5. Iterate | Continuous updates based on reviews and App Store data | High | AI assisted in fixing bugs and adding features |
Tool stack
- Claude 3.5: the core code generator. The non-technical founder described requirements in natural language; Claude understood the intent and produced Swift code.
- Cursor: the IDE. Made the “chat-to-code” loop smoother.
- Xiaohongshu: the core distribution channel. Chen built a personal IP on Xiaohongshu — product promotion at near-zero cost.
- App Store: the monetization channel. Paid downloads + in-app purchases.
Roles one person carried
| Role | AI replacement | Notes |
|---|---|---|
| Product manager | 0% | Demand discovery relied entirely on human insight |
| Development | 95% | Claude + Cursor produced nearly all the code |
| Design | 60% | AI generated the basic UI; human tuned aesthetics |
| Marketing | 10% | Xiaohongshu content marketing relied on human creativity and IP |
| Customer support | 30% | App Store review replies + community interaction |
3 Reproducible Key Moves
- Find needs in user communities, not in technology: Chen didn’t learn AI coding first and then hunt for something to build. He spotted a pain point on Xiaohongshu first, then used AI to build it. Spend 30 minutes a day browsing the community your target users live in, and log recurring complaints and unmet needs.
- The non-technical AI dev loop: requirement → natural-language spec → Claude implements → ship: You don’t need to learn to code, only to clearly describe “what I want.” Write what the app should do in Chinese, paste it into Claude, let Cursor generate it. When something breaks, describe the problem and let the AI fix it.
- Use content IP for free distribution: Continuously share your AI-build process on Xiaohongshu / Jike (the Chinese-language Build in Public), accumulating followers as your free distribution channel.
Limits and risks
- App Store competition is brutal: Fill-light apps have a low barrier; copycats will flood in fast.
- Revenue ceiling: A single app has limited monetization ceiling in China; you need diversified revenue.
- Technical debt: AI-generated code may accumulate problems over time, and a non-technical founder has limited ability to fix bugs.
- Platform dependency: Heavy reliance on the App Store and Xiaohongshu — rule changes can directly impact revenue.