A 22-year-old college student in Shenzhen made $14.7k in silence in March using a 900 RMB Redmi phone. His parents still think he's studying for exams in his room.

Every half hour, the phone automatically opens six Chinese news apps, scans the headlines, takes screenshots, and then sends the summaries to Claude. The AI compares the prices on Polymarket and directly provides betting suggestions.
He didn't write a single line of code throughout the process. It's all based on an open-source project called PhoneDriver — the principle is simple: AI looks at the screen like a human, judging where to tap and swipe on its own. You just need to clearly describe the task in plain language. The phone doesn't need to open any APIs or touch the app's underlying code; the model only sees the pixels on the screen.
The logic behind this student's approach isn't complicated. Chinese news is published in Chinese; any regulatory actions, trade decisions, or economic data, even before Reuters translates them, are already on platforms like Toutiao hours earlier. Meanwhile, the pricing on Polymarket is set by traders in the English-speaking world. When Chinese news is released and the market reacts, there's a lag of 20 to 50 minutes.
Among the six apps he uses, three block web crawlers — one blocks each. But PhoneDriver doesn't parse data; it just uses the app like a normal person, only faster and without breaks. 96% of mobile apps don't have public APIs, but every app has a screen. For PhoneDriver, that's enough.