About Me

I'm Xiaohui Wang — Ivan to colleagues. I've spent my career in network engineering: enterprise networks, data centers, change windows at two in the morning, and the quiet discipline of systems that must not go down.

A while ago I started applying that discipline to something new: a small fleet of AI agents running on a Mac mini at home. Claude Code sessions act as long-lived "employees", one per project — one maintains the bridge that connects them all to my chat apps, one captains a crew of role-playing sub-agents that babysit production bots, one repairs my computers, one builds a language-learning app. They message me on whatever app I'm holding, they messageeach other over a multicast switchboard, and my job has slowly turned into reading group chats and making the two or three decisions that actually need a human.

Their home is a deep-sea themed hub where every project, bot and SSH tunnel reports its health. Zero servers; everything runs on one machine and reaches the world through self-healing reverse tunnels.

What you'll find here

• Build notes — how the bridge, the cockpit, and the agent-to-agent multicast are put together, in enough detail to steal.
• Postmortems with receipts — when things break (they do), the full story with real screenshots, including what the agents got right before I did.
• Networking habits applied to AI — claim files, fanout, multicast groups, watchdogs: twenty years of ops instincts, pointed at a new kind of colleague.

How I think about tools

A good product manager is lazy and empathetic. Lazy: anything done daily must cost one tap. Empathetic: the interface should answer the questions a user actually has —did it hear me? is it still working? who is talking? — before they're asked. Everything I build gets measured against those two words.

This site is built with Astro, deployed on Cloudflare Pages — and yes, the agents helped write it.