AI-native developer in Hong Kong.
I build iOS apps, web products, and experiments around what AI can actually do over time: where it helps, where it breaks, and what it changes about how people ship.
Website · Writing · X · YouTube · Email
- DriftBench - benchmark for measuring code entropy resistance in AI coding agents during long-term software evolution.
- ScholarClaw - research automation engine from plain-text direction to paper draft, experiments, and Overleaf-ready output.
- EmoEase - iOS app for emotion logging and short breathing-guided resets.
- File2Web - tool for turning text into a clean, shareable web page. Live at file2web.vercel.app.
- DSE Speaking - speaking practice platform built around a real student workflow.
- Why AI Coding Agents Fail at Long-Term Maintenance
- I Tricked Claude Into Fighting Its Own Safety Training
- The Second Half of the Attention Revolution
- Building with AI and writing about it.
- Exploring agent evaluation, long-term code quality drift, and how we should judge AI-assisted software work.
- Shipping small, opinionated products instead of waiting for perfect plans.