There is a category of book written by extremely senior people about an emerging technology, and it has recognisable characteristics. It is short. It is written at a level of abstraction that never quite touches an example. It raises grand questions and declines to answer them.
It has the tone of a document circulated among people who already agree with each other. The Age of AI belongs to this category, and judging it fairly means being clear about what such a book can and cannot do. The distinctive contribution is the geopolitical framing, and I want to give it proper credit because it is genuinely underserved elsewhere. Most writing about artificial intelligence comes from technologists, ethicists or economists.
This comes from people who spent careers thinking about state power and international order, and that lens produces observations the other traditions miss. The treatment of AI as a strategic capability that changes the balance between nations, the discussion of how automated systems compress decision timelines in security contexts, and the question of what deterrence means when capabilities are opaque, are all worth thinking about and rarely raised this directly. The section on security and the speed of automated decision making is the strongest in the book. When systems act faster than humans can deliberate, the traditional apparatus of crisis management assumes conditions that no longer hold.
Kissinger's background makes this the part where the authors are genuinely expert, and it shows. The book is also short, which I count as a virtue. It makes its points and stops rather than padding a long essay into a volume. Now the substantial problems.
The book is abstract almost throughout. Claims are made at a level of generality that makes them impossible to evaluate. Artificial intelligence will change how humans understand reality. Perhaps.
In what specific way, with what mechanism, and what would count as evidence either way? The book rarely descends far enough to say. Reading it, I kept waiting for a concrete case study that never arrived. Without that grounding, the argument cannot be tested and the reader cannot do anything with it. The technical understanding is shallow.
The authors describe machine learning in broad terms and the descriptions are sometimes imprecise in ways that affect the conclusions drawn from them. A reader who knows the technology will notice places where a claim about what AI does is doing more work than the underlying reality supports. Given that one author has deep industry experience, this is more disappointing than it would be otherwise, and it suggests the book was written at a level of remove from the actual technology. The absence of conclusions is the most frustrating aspect.
These are people with genuine access and influence. If anyone were positioned to say what governments should actually do, it is these authors. What they offer instead is that the questions are profound and require serious thought. That is true and it is not enough.
A book that identifies an urgent problem and declines to propose a response has done half a job, and the half it skipped is the one their positions uniquely qualified them to attempt. The register is closer to pronouncement than argument. Claims arrive already settled rather than being built from evidence a reader could examine. That style works when the reader is inclined to defer to the authors' authority and works poorly otherwise, and it makes disagreement difficult in an unproductive way, since there is often no stated reasoning to disagree with.
The philosophical material on human self-understanding is where opinions will split most. It asks what it means for human reason to no longer be the only source of complex insight, and whether that changes how we understand ourselves. Those are real questions. The treatment is more evocative than rigorous, and readers who want philosophy done carefully will find it thin, while readers who simply want the question raised may find it valuable.
Two point nine reflects a book with one genuinely distinctive contribution, the geopolitical lens, wrapped in a great deal of abstraction that does not earn its place. It is short enough to be worth an evening if the strategic angle interests you. It is not a good introduction to artificial intelligence, it will not teach you anything technical, and its most important questions are raised and then left where they were found.