If you are reading this book at a quiet moment, I’m sorry. I mean: I wish you peace and quiet — but I must rob you of it before we can begin. For the problem at hand has a certain pace, and it is not dramatised. It is measured. One week in October 2026 is enough to illustrate it.
Five levels were moving simultaneously during those days. Not one after the other. Simultaneously. And that is the real finding: on none of the five levels did anything happen that was, in itself, unusual. What is unusual is simply that they moved together — and that no leadership layer, no manual, no market framework had an answer ready that did not come from a sales pitch.
Level 1: Technology
In October 2026, OpenAI demonstrated a Decisions API at its DevDay: a fast, inexpensive model that selects from short lists of options — yes or no, this or that, here or there. An industry observer summed it up in a single word: “basically, it’s Jev.” He was referring to an architectural principle that will play a central role in this book: controlling agents’ decisions not by reading all the content, but by evaluating options at the decision point. In 2026, the industry is building what governance architects had formulated as a requirement — only it is selling it as a feature, not as a matter of responsibility.
In the same month, Anthropic reduced the operating costs of its smallest model by around 75 per cent. Cutting prices in half also halves the margins from which security testing is funded. And cheaper tokens do not make agents cheaper overall — they make it possible to have more permanent agents for the same budget. The system grows with every price reduction. This is not a fear; it is the familiar logic of the Jevons paradox, applied to machines that never sleep.
And then there are the Dots: small, friendly companions that listen, join in the conversation, and stay by your side. Permanent agents for the home. We’ll come back to them later — not because of their technology, but because of what they reveal about the interface between convenience and control. For now, one thing suffices: they were bought before anyone had asked what they might be eavesdropping on.
Level 2: The Market
Also in October 2026, a company that one might have considered a reliable champion of sovereignty made the headlines: Ecosia, the search engine that plants trees, announced it was parting ways with Mistral. The reasoning was not dramatic, but rather a matter of careful consideration — and that is precisely why it is instructive: European sovereignty is not a guarantee that can be bought. It is a state that must be nurtured — and can be lost, even with European tools. Anyone who sees dependency solely as a US problem has failed to grasp the lesson: dependency is a structure, not a logo.
At the same time, the stack came of age. Mistral, n8n, self-hosted infrastructure, open weights: what had been a collection of disparate components a year ago began to take shape as a workable architecture. For the first time, the market provided everything an agent-based system needs — and once again left the most important question to the buyer: who controls it?
Level 3: Politics
In the US, the Federal Trade Commission summoned representatives from the agent industry; in the Senate, no senior executives appeared. In Europe, a Franco-German initiative emerged to treat AI exports and dependencies as an economic instrument — the digital supply chain, viewed through a political lens.
This is the level that executives are most likely to underestimate, because it feels so slow. It is not. The question ‘Where does your intelligence come from?’ has shifted this week from a procurement issue to a geopolitical one. Anyone who leaves it solely to the procurement department has already answered it — incorrectly.
Level 4: Civil society
The Wikimedia Foundation warned against the use of its projects by autonomous agents: bots that left behind masses of entries without the community behind them being able to understand the work. The UN Human Rights Council addressed the issue of constant surveillance via everyday devices. Citizens’ initiatives formed to oppose ‘always-on’ agents in private spaces.
Something happened here that I consider to be the real signal: the first entity to systematically ask about control was neither the state nor the market — it was the users and the communities. Civil society identified the key issue before the companies did.
Level 5: Europe and the discourse
In the Netherlands, government bodies piloted their own language model, GPT-NL. In Brussels, the deadlines for the EU AI Act’s obligations were postponed — the high-risk rules did not come into force earlier, but later. And on LinkedIn, the discourse was celebrating ‘Human Transformation’ — the very same words that this book takes seriously, as a content format to be posted that very afternoon.
The single insight I’d like to share with you here is this: this week, Europe is behaving like an organisation that has discovered the right question — but has not yet structured it. Twenty-seven islands, each with its own model, its own language, its own sense of urgency. What’s missing isn’t the will. It’s the semantics between the parts.
What this week means — and what it doesn’t
This week is not evidence of a conspiracy, nor is it hype or a catastrophe. It is evidence of speed without junctions. Five levels are in motion, and no one controls the interfaces between them. Technology delivers, the market delivers, politics reacts, civil society takes note — but the question that affects all five levels in common is not answered anywhere, because it is not asked anywhere:
Who exactly is at the helm here?
This book is about to slow down — from the next page onwards. It will take the five-level acceleration as a given and examine a single architecture: the answer that is the same across all five levels. It is not technical. It is not expensive. But it is inconvenient, because it requires a person.
This book is about junctions. And about the person standing at them.
