AI Agent Deep Dive is an educational resource and minimal Python agent implementation that teaches the core architecture of AI agents through source code analysis and a teaching project.
The repository addresses the need to understand how AI agents are structured internally by providing two main components: deep-dive analysis documents examining the source code of existing agent systems, and a minimal teaching implementation that demonstrates how an agent's core components should be organized. The teaching project focuses on clarity and simplicity, concentrating on the agent main loop, a replaceable LLM interface, skill discovery, and CLI scaffolding without unnecessary engineering complexity.
The teaching agent is designed for developers learning agent architecture rather than for production use. It uses a fake LLM interface that echoes user input as streaming text responses, allowing learners to understand the agent structure before integrating real model APIs. The core code is intentionally concentrated in a small scope, with the agent implementation in a single file and supporting documentation provided. This approach makes it suitable for educational purposes and as a foundation for understanding how to swap in real language model integrations.
The project maintains analysis materials focused on learning and commentary rather than providing full source code directories. The teaching implementation uses Poetry for dependency management and includes a CLI entry point for running the minimal agent. Documentation is provided in the docs directory alongside the core agent implementation.
