Mark Zuckerberg's AI manifesto dropped this week and it is, by any measure, a maximalist document. He envisions AI as humanity's intellectual companion, a co-pilot for civilization. What the document elides, with remarkable consistency, is the question of desire. Who decides what superintelligence optimizes for? Whose preferences get weighted?
The Agentic AI Enclosure Problem
A timely 2026 paper on arXiv by Gamba, Romero, and Schoenebeck, Agentic AI: User Empowerment or Enclosure?, frames the tension precisely. Agentic AI systems promise to act on users' behalf, but the paper finds that the structural incentives of platform companies push toward enclosure rather than empowerment. The agent serves the platform that trained it. Zuckerberg's manifesto is, under this lens, not a vision of human flourishing. It is a remarkably candid description of the world's most sophisticated enclosure apparatus.
Taste Cannot Be Abstracted
This is where the New Yorker's dissection of Leopold Aschenbrenner's hedge fund implosion becomes relevant. Aschenbrenner bet on AI hype as investable reality. The market said: not yet, not like that. The same failure of abstraction haunts Zuckerberg's document. Superintelligence is legible. What a specific human in a specific mood wants on a Tuesday afternoon is not. Culture Slop's conversation with designer Soleio is instructive here: his argument that a total addressable market of one is sufficient is the photographic negative of Zuckerberg's manifesto. One is a vision of scale. The other is a defense of particularity. The AI bubble inflates because most investors, and most CEOs, can only see the first.