Skip to content

Backstory: How This Book Came to Be

The Backstory

While reading through the Agentic Design Patterns textbook, I got curious about real-world implementations. The patterns were fascinating in theory, but I wanted to see them in action.

I'd heard about OpenAI's Codex CLI - a production AI coding assistant that had been making waves. Perfect, I thought. Let me dive into their codebase to see these patterns in practice.

Plot twist: The entire codebase was in Rust. 🦀

Not being fluent in Rust, I did what any pragmatic developer would do - I enlisted an AI assistant (Cursor) to help me analyze the codebase and extract the agentic patterns. What started as a simple exploration turned into something much bigger.

The Process

Using Cursor, I systematically went through Codex's ~100,000 lines of Rust code, identifying real implementations of:

  • Prompt Chaining in their conversation management
  • Tool Use in their MCP integration
  • Exception Handling in their retry mechanisms
  • Human-in-the-Loop in their approval workflows
  • Sandbox Escalation in their security systems
  • And 12+ more patterns...

Each pattern I found was then abstracted into clean Python implementations, complete with explanations of how Codex actually uses them in production.

What Makes This Different

This isn't just another AI tutorial. It's a grounded textbook that bridges the gap between:

  • 📚 Theory (the original Agentic Design Patterns book)
  • 🏭 Production Reality (OpenAI's battle-tested Codex implementation)
  • 🐍 Practical Code (Python examples you can run and modify)

Every pattern here is backed by real code from a system handling millions of interactions. No toy examples, no academic hypotheticals - just production patterns you can actually use.

The Tools Behind This

Why Open Source?

The original textbook and Codex are both open resources. This felt like a natural extension - taking those ideas and making them more accessible to Python developers who want to build production-grade AI agents.

Plus, the best way to learn is by doing. All code is MIT licensed, so fork it, break it, improve it, and share what you learn.

Ready to dive in? Start with Chapter 1: Prompt Chaining or jump to any pattern that interests you.

Discussed on Hacker News