August 2026 became a bifurcation point in the public discourse on artificial intelligence. Three figures shaping the mainstream of technological thought published texts that paint three incompatible pictures of the future. Yet on one point they are united: the technology is already delivering returns. The question is merely who pays for it and why the author's role is increasingly reduced to that of a verifier.
A Schism in Visions of the Future
Mark Zuckerberg, in his essay The Future is for Everyone, promises personal superintelligence to every inhabitant of the planet, betting on expanding human capabilities through private agents and compute auctions. Bill Gates, by contrast, begins with a troubling admission: no plan exists for entering the new era. He proposes a "reserve" for professions, a tax on AI tokens and robots, as well as new institutions to manage the transition. Between them stands consultant Nikhil Suresh, whose firm spent a year and a half studying corporate AI projects from the inside and recorded a zero rate of successful implementations.
Illustrative is the case of a top manager at a company with revenue exceeding $2 billion, who presented a strategy built around AI without ever having opened a single AI tool. Such admissions, according to Suresh, carry the risk of dismissal. This is not merely an anecdote but a systemic coordination problem: public confirmation of effectiveness preserves one's position, while honest assessment destroys a career.
The Economics of the "Reverse Centaur"
The key phenomenon I observe in the current market dynamics is the inversion of the "centaur" model from chess in 1997. Back then, the human was the head, and the machine was the body. Today, the algorithm makes decisions, while the human performs physical or cognitive work at the machine's pace, verifying results. A programmer with GitHub Copilot, a lawyer signing a document prepared by a model, a driver under recognition cameras—all become "reverse centaurs."
The economics here are harsh: verifying each result costs almost as much as producing it. Savings arise only because no one pays for the hours spent on verification. According to the Stanford Digital Economy Lab, employment among young professionals aged 22–25 in occupations highly exposed to AI optimization is 19% lower than among their peers in less affected fields—and this gap has grown from 15% over the past year. A verifier is needed only as long as the error rate justifies their salary. Eight errors per hundred cover the costs; one in ten thousand no longer does.
The Time of the Intermediate State
Block's 20% stock rise in February after cutting 4,000 employees, and the 5% decline in August following excellent earnings, demonstrate that the market punishes uncertainty about returns on AI investments. Companies adopt the technology faster than its effectiveness is confirmed, and the exchange reacts instantly to announcements of layoffs, while real figures only appear a quarter or two later.
Billionaires argue over who will get the technology, but a far more fundamental question is being decided—who will submit to it. So far, none of the proposed mechanisms—neither Gates's "reserve" for professions nor Zuckerberg's compute auctions—provides an answer to the main challenge: how to preserve the worker's right to decide when and how to deploy an AI tool. Without this, any talk of "empowerment" remains mere rhetoric concealing a new form of hierarchical control.
My assessment: we are in a phase I call the "verification window." It will close as soon as models reach the threshold of error-free performance, and then millions of "controllers" will become redundant. The question is not whether that moment will arrive, but whether societies will manage to create the institutions that Gates rightly calls missing. For now, the only rational advice for professionals is not to invest in checking others' results, but to develop the skill of discerning where the machine is appropriate and where it is not.