The Architect Who Drew the Flood, Then Stood in It

Three years is not long. It is long enough. When Mustafa Suleyman and Michael Bhaskar published *The Coming Wave* in 2023, the book read as an urgent warning from a man with his hand on the lever. By 2025, Suleyman was CEO of Microsoft AI, one of the most powerful positions in the industry he had spent 300 pages arguing needed containment. The book has not changed. Its author's coordinates have. This is the kind of irony the book itself might have predicted—it spends considerable energy explaining why the incentive structures driving technological development are essentially irresistible—but it never quite reckons with the possibility that its own author would become a primary vector of the wave he named. The ten steps toward containment outlined in Chapter 14 now read less like a policy blueprint and more like a letter of intent filed before a merger. Some of the steps have been attempted. None have been achieved at the scale the book demands. The EU AI Act, which Suleyman critiqued as too narrow, has stumbled into enforcement with roughly the limitations he anticipated. China's regulatory apparatus, which the book credited with surprising proactivity, has oscillated between crackdown and acceleration depending on the quarter. The geopolitical AI race the book described has intensified beyond even its own projections—not because the analysis was wrong, but because the timeline was generous.

What the book got right is substantial and specific. The prediction that large language models would become ubiquitous general-purpose tools, that synthetic biology would accelerate past regulatory capacity, that autonomous drones would reshape asymmetric warfare—all of this has materialized with uncomfortable precision. The Ukraine chapter, which used the Russian invasion as a case study for technology-enabled asymmetric resistance, now looks like a modest preview of what drone warfare has become across multiple theaters. The discussion of deepfakes and AI-driven disinformation anticipated the 2024 election cycles in the US, India, and elsewhere with depressing accuracy. The concept of "fragility amplifiers"—technologies that don't cause crises but make existing systems more brittle—has proved to be one of the book's most durable contributions. Every major cybersecurity incident since publication has essentially illustrated this chapter. Where the book stumbles is in its treatment of AI capabilities themselves. Written in the GPT-3.5/GPT-4 transition period, it hedges carefully on timelines for artificial general intelligence while simultaneously arguing that the rate of improvement is "well beyond exponential." The hedging was wise. The framing was not. What has actually happened is messier: capabilities have surged in some domains and plateaued in others, the scaling laws have shown diminishing returns in certain architectures, and the definition of "general intelligence" has become even more contested. The book's tendency to treat AI progress as a single ascending curve now looks like the bias of someone who spent a decade inside the most successful AI lab in history.

The deeper blind spot is structural. Suleyman writes as though the primary tension is between technological capability and governmental capacity—the nation-state versus the wave. This framing inherits something from Kaczynski's technological determinism (a debt the book would not acknowledge) and something from Kai-Fu Lee's geopolitical framing of AI competition. What it misses, or at least underweights, is the degree to which the wave is shaped by capital allocation, not just invention. The book discusses incentives, yes, but treats them as atmospheric conditions rather than as decisions made by identifiable people in identifiable boardrooms—boardrooms the author now occupies. The absence of any serious engagement with labor movements, with the Global South's position as both data source and market, or with the environmental costs of training and running frontier models at scale is conspicuous in 2026. The book's vision of containment is top-down: governments, treaties, technical safety research, industry self-regulation. The possibility that containment might also look like organized refusal, like communities rejecting specific deployments, like workers demanding a seat at the table—this does not appear. It is a book about power that does not fully interrogate who holds it.

What hits differently now is the prologue, written by an AI, which Suleyman included as a kind of parlor trick to demonstrate capability. In 2023 it was a novelty. In 2026 it reads as a timestamp—a record of what AI prose sounded like at a particular moment, already dated in its cadences, already surpassed in fluency. The gesture was meant to unsettle. Now it merely dates. More striking is the "grand bargain" chapter, which argues that the nation-state's legitimacy depends on its ability to deliver safety and prosperity in the face of technological disruption. Since publication, we have watched multiple democracies struggle with exactly this bargain, not hypothetically but in real time, as AI-driven job displacement has begun to register in employment statistics rather than think-tank projections. The book sits at a hinge point in the corpus: it synthesizes the speculative anxieties of earlier works—Dick's questions about machine consciousness, Bacigalupi's biotech dystopias, Gibson's corporate power structures—and attempts to translate them into policy language. It is the moment when science fiction's warnings get a McKinsey deck.

The question the book now raises, which it could not have raised when written: What happens to a containment argument when its author becomes the thing to be contained?