Build an AI Coding Agent

A model asks to read a file. What happens next?

Something must validate the request, check permissions, execute the operation, preserve its result, and decide whether the model should continue. That system is the agent harness.

Build an AI Coding Agent teaches you how to build and understand it.

Through 18 chapters and 125 quiz questions, you will follow a practical learning sequence: understand a concept, predict what will happen, inspect the implementation, and investigate a failure. Separate solutions explain the reasoning behind each answer.

You will learn how to:

• Build an agent loop that connects model responses, tool calls, and conversation history.
• Define tool contracts, validate arguments, and enforce execution permissions.
• Handle timeouts, cancellation, retries, and uncertain outcomes.
• Save checkpoints and recover after a crash without blindly repeating actions.
• Manage context, streaming, concurrency, configuration, and secrets.
• Use traces and evaluations to investigate behavior.
• Build a responsive terminal interface.

The concepts apply across programming languages. The working case study, CaptainCode, uses Go, OpenRouter, SQLite, and Bubble Tea. A pinned source revision connects explanations to inspectable code, while deterministic exercises let you begin without a model account.

This book is for developers comfortable with functions, APIs, files, and Git. Previous experience building AI agents is unnecessary. A Go reading guide helps you follow the implementation.

Written from the perspective of a practicing systems engineer and mentor, the book emphasizes clear reasoning, observable behavior, and honest implementation limits.