In recent months, AI has been making headlines again, even as large organizations invest more than ever to deploy agents. Last summer’s worry—will AI really have an impact? has given way to a more existential fear: does AI have too much impact? The restrictions impo

The restrictions imposed by U.S. authorities at the launch of Mythos 5, Fable 5, and ChatGPT-5.6 served as a wake-up call, reinforced by OpenAI’s announcements in July about a cyberattack caused by a model in testing that escaped its “training cage.” The question of artificial general intelligence (AGI) is back on the table, along with its corollaries—the end of work? Of freedom? Of humanity? complicating the job of executives who are being told to rally their organizations around an AI transformation plan.

But what are we actually talking about? AGI refers to artificial intelligence capable of performing every productive mental task at least as well as the (best) humans. In short, “a country of geniuses in a data center,” in the words of Dario Amodei, CEO of Anthropic. While some voices have claimed that this ambition is already all but a reality, most researchers (76% in a 2025 survey conducted by the Association for the Advancement of Artificial Intelligence) believe further scientific breakthroughs will be needed to get there.

And where do humans fit into all this? They’re holding up well. AI researchers speak of the “jaggedness” of current AI capabilities—far better than humans at some skills, far worse at others. Even in software development, where the frontier has shifted considerably in AI’s favor, humans and their judgment remain critical for architecture decisions, trade-offs on features, and the allocation of responsibilities.

And in three, five, ten years? Who knows ? The science falls short: Our learning models work when a decision is either right or wrong, not in the gray zone that nonetheless makes up a good part of our daily lives. Learning on the job, through apprenticeship, is the foundation of our organizations. With all its inconsistencies and ambiguities, it does not lend itself well to codification. But the AI labs are working flat out on the next generations of models. We are running short on data—“there is only one Internet,” in the words of Ilya Sutskever, one of OpenAI’s co-founders. But physical AI and “real-world” models will expand the playing field. We may be running short on energy and raw materials to sustain exponential demand for computing power. But here, too, innovation is moving quickly...

What executive doesn’t occasionally feel a touch of vertigo when contemplating the effort required to align a company with today’s AI, with the more or less distant prospect of AGI looming? On this subject, Blaise Pascal remains the best consultant of all. The philosopher was, to be sure, “frightened by the eternal silence of these infinite spaces.” But he had an antidote in mind: the wager, whose executive version goes like this. If I transform my organization today and AGI is slow to arrive or never comes at all, I win; my efforts were not in vain. If AGI does eventually arrive, I end up in the same position as those who did not transform. For any good gambler, there is only one rational decision: transform. Place your bets!