Harari's most lasting contribution is not what he predicted. It is what he named — a condition that was already forming and that most people not yet found words for.

There is a particular kind of intellectual work that consists not of solving a problem but of making it visible — finding the name for something that was already happening but had not yet been articulated clearly enough to think about properly. Homo Deus is that kind of book. Its speculative passages — the predictions about immortality, the god-like ambitions of biotechnology, the emergence of a post-human elite — are the parts most likely to date, and some already have in ways that are instructive. But the concept at the book's centre, dataism, has not dated. It has, if anything, become more precisely descriptive of the world it was trying to anticipate.

Harari defines dataism as an emerging ideology that places data and its processing at the centre of meaning and decision-making — not as a tool in service of human judgment but as the organising principle that increasingly replaces it. The shift he is describing is not just technological, but philosophical: a slow displacement of humanism, which locates value and authority in individual experience and autonomous choice, by a system that locates them in data flows and algorithmic processing. When an algorithm can predict your preferences more accurately than you can articulate them, when institutional decisions — credit, employment, medical treatment — are increasingly delegated to systems that operate below the threshold of human legibility, the question of where meaning resides becomes genuinely unstable.

Two of Harari's more specific predictions are worth exploring further to see what they reveal about the underlying argument. The first is the "useless class" — a segment of society made economically redundant not by poverty but by the obsolescence of their cognitive and physical labour in relation to what automated systems can produce more efficiently. The prediction was received in 2015 as provocation. What has happened since is neither its simple confirmation nor its refutation but something more complex: the redundancy has proceeded unevenly, in ways that expose its dependence on political choices as much as technological inevitability. Which labour becomes redundant, on what timeline, with what consequences, and for whom — these turn out to be questions that dataism itself is poorly equipped to answer, precisely because they require value judgments that algorithms can optimise for but cannot make. The useless class, it turns out, is partly a technical phenomenon and partly a political decision about whose experience counts as a cost worth bearing.

The second prediction — that humanism as an organising ideology will be displaced by dataism — has fared differently. Here Harari was describing a process rather than an event, and the process has continued with a consistency that the book anticipated more accurately than most contemporary commentary. The language of data-driven decision-making, of evidence-based policy, of algorithmic personalisation — these have become not just technical but normative prescriptions, carrying the implicit claim that what is measurable is what is real. That is dataism operating as ideology rather than method, and it was already fully formed in the corporate and institutional cultures Harari was observing. What he added was a name and a historical frame.

The trajectory of the work that followed Homo Deus is itself worth noting. Sapiens had asked open questions about the stories humans tell themselves to organise collective life. Homo Deus extended that inquiry into genuinely uncertain territory. The work that followed moved progressively toward a different register — more prescriptive, more confident, more oriented toward answers than toward the productive difficulty of the questions. It is an observation about what happens to ideas when they achieve the kind of cultural velocity Harari's did — when the demand for clarity and applicability begins to shape what the thinking is allowed to do. Which is, in a different register, precisely the dynamic dataism describes: the pressure to produce outputs that are easy to understand, scale, and consume.

The methodological stakes of all this are specific. An analyst working alongside AI systems — using them to process a world increasingly organised around what they can measure — faces a structural problem that the technology does not solve and tends to obscure. The instruments are not neutral. The categories built into any system of data collection are prior decisions about what counts as evidence, what counts as a meaningful outcome, and whose experience is relevant to the system. What gets called AI today covers an enormous range of genuinely distinct capabilities and limitations, but what unites them is that they optimise for what they have been told to optimise for. Choosing what that is — and understanding what is lost in the choosing — is not a technical question. It is the central analytical question, and it remains stubbornly human.