Building AI Agents

Build practical AI agents - without the hype.

AI agents are everywhere in today's tech conversations, but many explanations are vague, overly ambitious, or buried in abstraction.

This book takes a practice-first approach.

You will learn how practical AI agents work by building around a real Python project: a small EXIF privacy agent that reviews image metadata, decides when cleanup may be needed, asks for approval before file changes, and summarizes results in plain language.

Instead of chasing vague autonomy, this guide focuses on the parts that actually matter: goals, planning, tools, safety boundaries, approvals, structured results, and feedback loops.

• What an AI agent is - and what it is not
• How agents differ from tools, workflows, chatbots, and MCP servers
• How to structure a small goal-driven agent loop in Python
• How to turn plain-language intent into structured plans
• How to use tools and capability layers safely
• How to keep read-only actions separate from mutating actions
• How approval gates and human-in-the-loop design make agents more trustworthy
• How structured tool results improve agent behavior
• How to test, evaluate, trace, and debug practical agent workflows
• How to add model-backed planning and ReAct-style reasoning without giving up safety boundaries

This book is centred on exif-privacy-agent, a practical Python agent that can:

• Review one image or a folder before sharing
• Detect privacy-sensitive metadata such as GPS fields
• Decide when cleanup may be useful
• Ask for approval before writing files
• Run cleanup workflows through a clear safety boundary
• Use either a direct local adapter or an MCP stdio capability layer
• Summarize results for humans while preserving structured outputs

This guide is for:

• Developers and builders exploring AI agents
• Technical creators who prefer learning by doing
• AI builders who want practical agent patterns without hype
• Readers who want a clear companion to an MCP server project
• Anyone trying to understand how planning, tools, approvals, and safety fit together

This is not a research-heavy survey of autonomous systems.
It is not a book about agent swarms or vague AI magic.

It is a focused, hands-on guide to building practical AI agents that are small enough to understand, useful enough to run, and safe enough to trust.