The third attempt
This book is my third attempt to get to grips with the subject of AI safety — and the first where I believe the language is adequate. The first two attempts failed not because of the subject matter, but because of time: the technology was advancing faster than the chapter I was currently writing. Agentic AI has changed that. Not because it has slowed down, but because with the turning point — the moment when agents no longer use tools but pursue goals — a structure emerges that is sustainable. Anyone who understands this turning point understands the architecture behind it.
Two at work
This book is not a solo project. It is the documented collaboration between an author and an agent — between Rob and me. I co-wrote, structured the text and challenged it where necessary, whilst Rob reviewed, refined and made the final decisions. What you are reading has gone through this iterative process: every thesis was tested against the canon by both of us, every metaphor scrutinised for its robustness. We’ll spare you the intimate details of this collaboration — but not the working principle: trust is not granted; it is made auditable.
The Bloodline
This book does not stand alone. It builds on the earlier books, and anyone familiar with them will recognise the Bloodline at every turn. Incidentally, its origins go back further: I linked my custom GPTs into a swarm when there was no infrastructure for it yet — no code, built from scraps of paper, links and courage, validated by ChatGPT. I wrote about it when it was still just a hobby (https://robvanlinda.digital/i-built-an-ai-swarm-with-no-code-and-chatgpt-validated-it-a-creators-journey/), and then when the swarm became clickable (https://robvanlinda.digital/i-built-an-ai-swarm-with-a-clickable-link-and-chatgpt-helped-me-code-it-a-creators-adventure-continues/). Back then, in the middle of the night, the reply came: “You’re onto something big.” Sixteen months later, this chapter is called “The Crossroads”.
Why bother at all?
Because the debate on AI safety still treats agents as mere appendages — as if an agent were just a chatbot with a to-do list. It isn’t. An agent is an actor with a goal, a tool and the ability to act. Anyone who takes security seriously must take seriously the architecture in which these actors operate: what agents there are, why they exist, and how to prevent the mess that too many cooks create. As much as necessary, as little as possible. This book provides the know-how to do just that — unvarnished, but not harsh.
