Architecture

My kid cracked the code on AI agents yesterday. Accidentally, obviously. I was practicing a presentation about agentic workflows, you know, pacing around the living room, narrating slides to nobody, and he was watching from the couch like I was performing a particularly confusing magic trick. The next morning, on the walk to school, he hit me with: “So what are those agent things, actually?” I did what any self-respecting engineer would do. I gave him a five-minute answer to a ten-second question. Context windows. Tool calling. ReAct loops. The whole parade. ...

OpenClaw (formerly Clawdbot / Moltbot) has been getting a lot of attention recently. People describe it as an “AI agent”, a “local assistant”, or even “an AI that controls your computer”. All of those are partially true — but they can also be confusing. (If you want to understand the basics of how AI agents work in general, I wrote a primer on that.) This article explains OpenClaw’s architecture at a high level, without diving into implementation details. The goal is to give you a clear mental model of what OpenClaw is made of and how the pieces fit together. ...