Preface

You do not need to know how to code, train a model or build an AI system to read this book. If you have used ChatGPT, Claude or another AI assistant, you already have enough experience to begin.

You do not need to know how to code, train a model or build an AI system to read this book. If you have used ChatGPT, Claude or another AI assistant, you already have enough experience to begin.

Maybe you asked one to help write an email. Maybe you handed it a long document and asked what mattered in it. Maybe you were simply curious and typed a question to see what would come back.

If you have done any of that, you have already used one of the most consequential inventions in modern artificial intelligence: a large language model, often shortened to LLM. These are systems trained on enormous amounts of text, able to follow language and produce responses that read naturally. They write, translate, summarise and answer hard technical questions.

But there is a point most people miss, and it is the point this book is built on. A language model, on its own, does not do anything.

It waits for text and returns text. It does not open your calendar. It does not send an email. It does not check today's flight prices. It does not form a plan and then carry it out. Everything a modern AI assistant appears to do comes from software built around the model: an interface, search tools, somewhere to store what it learns about you, and logic deciding when each of those gets used. The model is the reasoning engine inside that system. It is not the system.

When such a system is built to pursue a goal rather than merely reply, we call it an AI agent. And that leads to the question this book exists to answer:

What happens when AI stops only answering, and begins working through steps — sometimes with other AI agents beside it?

What this book covers

The journey runs in one continuous line. We start with the language model itself and what it is actually doing when it produces text. We give it better instructions, then access to current knowledge. We wrap it in tools and memory until it can pursue a goal, which is the moment it becomes an agent. We look inside the assistants you already use, ChatGPT and Claude, to see what happens when one of them searches or works through several steps for you. Then we reach the newer territory: several AI agents dividing work between them, what that buys, what it costs, and where a person still belongs.

Three things get sustained attention because they are where the interesting questions now live.

• A real system, examined closely. Chapter 11 walks through how Claude handles deep research and how it handles writing software, using the engineering accounts published by the team that built it. The two jobs use different designs, and the reason why is one of the most useful ideas in the book.
• AI agents working together. Part V covers when to move from one agent to several, four simple ways they can cooperate, and how to build such a system without it collapsing under its own complexity.
• How you know it is ready. Part VI asks the question that decides whether any of this reaches real users: how do you know an agent behaves well enough to be trusted with the authority you are about to give it?

A note on Volume 1

This is the second volume of AI Agents for Everyone, and it is written to be read on its own. Nothing here depends on remembering the first book. Every idea the later chapters need is built up from the beginning, in this volume, in order. If you have read Volume 1 you will recognise the foundations and find them expanded; if you have not, you will not notice anything missing.

What you will be able to do

• Explain how generative AI and large language models work, at a level you can use.
• Write instructions that get better results, and understand why they do.
• Say what retrieval adds to an AI system, and what it does not.
• Tell the difference between a chat window and an agentic system, precisely.
• Design an agent deliberately, and choose its tools and permissions on purpose.
• Recognise when a problem needs several agents, and when one agent is the better answer.
• Read a real production design and understand the trade-offs its builders made.
• Ask, and answer, whether an agent is ready to be trusted with a decision.

Who it is for

Anyone who wants to understand these systems without getting lost in vocabulary. That includes students from any discipline, professionals and executives weighing where agents fit in their work, entrepreneurs, educators, developers new to agentic AI, and readers who are simply curious. No prior experience in AI or machine learning is required.

How to read it

• New to the subject? Start at Part I and read straight through. Each part answers a question the previous one raised.
• Already comfortable with language models and prompting? Skim Part I, then settle in at Part II where agents begin.
• Here for the multi-agent material? Chapter 11 and Part V are the centre of the book, and Chapter 5 gives you the vocabulary they assume.
• Building something now? Parts III and V, then Part VI before you put it in front of anyone.

Any code in this book is optional and clearly marked. A reader who skips every listing will still understand the whole argument.

A word before we begin

AI is no longer a future concept. It is already shaping industries, organisations and ordinary working days. Agents are the next step in that shift: systems that plan, use tools and carry out work with some degree of independence.

My own view is that they will become unremarkable, the way spreadsheets and email did, and that the people who benefit most will be the ones who understand what these systems are actually doing. Not the ones who can build them from scratch. The ones who can look at a problem and say what the software should be allowed to handle, what it should hand back, and how anyone would know it was working.

That is what the next few hundred pages are for. Let us start where the technology starts.