Build real things
with AI agents.

What AI Agents Are (and Aren't)

An AI agent is a piece of software that can take actions on its own — write code, create files, run commands — not just answer questions. But agents are narrow: they do what you tell them, eagerly and literally. Your job isn't to do the work. It's to direct it clearly.

Memory and Context

AI agents have limited memory. Every time you start a new session, the agent starts fresh — it doesn't remember what you did before. To work well with agents, you need to think about how to give them context and how to get them to leave notes for next time.

Tools of the Trade

There are five ways to work with AI: talk to it in a chat, use a platform that builds and hosts software for you, deploy enterprise agents that handle your customers, connect your apps with workflow automation, or use an agent workspace where you control the files and decisions. This chapter maps the landscape.

Setting Up

You need three things: VS Code (free), the Claude Code extension (free), and a Claude subscription ($20/mo). This chapter walks through every step — downloading, installing, and running your first command.

Git and GitHub

Git is an infinite undo button for your project. GitHub is where your code lives — the canonical location that gives you backup, version history, and portability. Together, they let you save your work, go back to any previous version, and move between machines or tools without losing anything. Your agent handles most of the mechanics — but you need to understand what's happening.

Configuring Your Agent

Claude Code reads configuration files to know how to behave. The instruction file (CLAUDE.md) is where you start. Rules, memory, and skills give it context. Agents, commands, hooks, and MCP servers give it capabilities. You don't have to create any of these files by hand — just ask Claude to set them up for you.

The Meta-Process

Every agent project follows the same shape: user input, orchestrator, explorer, actor, reviewer, iterate. Add a post-mortem and a change process, and the system gets better every time you use it. This chapter defines the framework. Chapter 8 shows it in action.

Content Writing Example

This chapter puts the meta-process from Chapter 7 into practice with a real project — a blog about flags. One command produces a full draft, using two agents, two skills, and a command file. Then we walk through how the system was built, step by step.

Writing Effective Instructions

Skills, agents, and commands are just markdown files — but how you write them matters. This chapter covers the craft: conditional logic, sub-agent dispatching, orchestration, blocking gates, skill sizing, skill composition, and using files as external memory.

OpenClaw

Everything so far has been about agents inside a code editor. OpenClaw is something different — a personal AI agent that runs on your own hardware, communicates through messaging apps, and operates 24/7 without waiting for your commands. This chapter covers what that world looks like, what it can do, and why you should be careful.

Building Software with Agents

The same process from Chapter 7 — orchestrator, explorer, actor, reviewer, iterate — scales to building software. The blog used two agents (researcher and writer). Software adds more: a manager dispatches, an explorer researches the codebase, a developer writes code, and a reviewer checks quality. Gates pause the process for your approval.

Claude Built-in Capabilities

Claude Code ships with built-in slash commands that manage your sessions, context, settings, and tools. These work out of the box — no setup required. Chapter 6 covered the configuration layer (rules, skills, agents, commands you create). This chapter covers the operational layer — the default commands you'll use every day.

Roy's Claude Config

Chapter 11 showed the software development process. This chapter shows one way to build it. I share my own system — agents, commands, rules, and project structure — as a reference. It's probably over-engineered for most cases, but it shows what a mature setup looks like. Take what's useful, skip what isn't, and build your own.

Servers, Hosting, and Deployment

Building something is half the job. The other half is putting it somewhere people can access it. This chapter explains what servers are, where your project can live, and how to get it from your computer to the internet.