Agentic AI Engineering
A hands-on repository of runnable tutorials that build an AI agent from first principles — the agent loop, tool executor, memory layer and evaluation harness — before introducing any framework.
Agentic AI Engineering is a tutorial repository for engineers who want to understand what underlies agents like Claude Code, Codex and GitHub Copilot by building the pieces themselves: the agent loop, tool calling, the memory layer and an evaluation harness, all before any framework is introduced. Each lesson is runnable in one command with uv run --directory python <lesson>.py, and a companion Substack series, "Disassembling AI Agents," reverse-engineers real production agents.
The material is presented as a path from a first LLM API call through to a production evaluation harness, aimed at developers who want first-principles understanding of tool calls and inference loops rather than only framework usage.