Building an AI Agent from Scratch in Go

What does it take to build an AI agent from scratch in Go — no frameworks, no magic?

In this talk we'll find out, taking inspiration from two real systems: Claude Code and OpenClaw. We'll break the problem into two halves.
The first is the agent core: context management (keeping conversations coherent when the context window is finite), tools (exposing Go functions as callable actions), skills (a Markdown-based plugin system with lazy loading), and subagents (delegating work to isolated agent sessions).
The second is the long-running agent: how to turn an agent into a daemon, add a heartbeat for proactive wake-ups, and schedule autonomous work with cron.
The second half of the talk is a live demo on a real running agent, showing all of these pieces working together.
You'll leave with a clear mental model of how agents work under the hood and concrete Go patterns you can use right away.