About

About Sun Xin

Sun Xin — AI Agent product manager, based in Shanghai. I work on designing enterprise AI agents and agentic workflows, and I argue for using benchmarks and evals to turn "what makes a good product" from a gut call into a standard a machine can actually check.

How I think about products

Some of it I've written up as long essays — for example "What an AI-era PM actually does" and "Why agent-era interfaces run backwards". More on the blog.

What I'm working on

  • Product manager at a startup building AI products for cross-border trade, where my day job is building the eval system for ToB agents;
  • I open-sourced Chinese-Geo — an AI-visibility audit for sites that need to be found and cited across China’s retrieval ecosystems (published on PyPI);
  • I open-sourced loop-engineering-skill — a verifier-first Claude Code skill that stops weak judges from wasting tokens on polished garbage, then designs the verifier and gates correctly;
  • Master's research: designing benchmarks for self-evolving agents;
  • On the side, I write AI skills for galgames.

The four look unrelated, but underneath they're the same thing: turning a fuzzy standard of judgment into executable instructions and measurable metrics.

Credo

Do the right thing — and make it beautiful.

Contact & links