Ideas is where I write from real operating problems.
Most pieces start with something concrete that is not working, then trace the system around it: what the incentives are doing, where judgment is being crowded out, how AI changes leverage, and what a simpler operating model might make possible.
How I Architected My OpenClaw Instance
The open-source OpenClaw runtime gives me the substrate. The real work was designing the operating layer on top: routing, memory, capabilities, state, lanes, and context discipline.
AI 需要经、法、术、器，不只是更强模型
AI changes leverage, not the logic of judgment. Chinese-first post with a separate English companion version.
Singapore Did Not Lose Agency Overnight
A systems essay on how centralisation can narrow local judgment while preserving state capacity.
The Model Is Not the Product. The Harness Is.
The real power in AI is not the model. It is the harness around it.