Meta's Chief AI Officer Alexandr Wang: Perception is More Real than Reality, People Choose to Believe Their Own Perception
Meta's Chief Artificial Intelligence Officer Alexandr Wang stated in a public talk that technology founders assume that better products and prettier data will help customers recognize value, but this assumption does not hold true for large companies and government clients. He specifically mentioned that the buyers are not startup teams, but large organizations and government systems.
Wang's original words were that many companies founders have worked with seem very data-driven, as if the truth would automatically reach everyone and form a shared reality; he believes that most large companies do not operate this way, nor do government entities. He continued by saying that for many large clients, perception is more real than reality because reality is often too unappealing, and few are willing to face it; people choose to believe their own perception.
He thus proposed a transaction mechanism: if the buyer is not judging reality, having a better product in reality is not enough to win the order; the orders flow to the party that controls perception. For enterprise sales and products, he divided the work into two halves: one half is improving reality, and the other half is shaping perception, with both having equal weight.
He specifically named Palantir, stating that its superpower is its ability to shape perception better than most tech companies, describing the company as a combination of a theater troupe and a software company. He added a specific practice: Palantir used to give all new employees a book on performance, and this practice lasted for a very long time. Public reports indicate that this book, which has long appeared on the onboarding reading list, is Keith Johnstone's "Impro," which discusses improvisation and drama.
Wang himself is the Chief AI Officer at Meta and is responsible for Meta Superintelligence Labs. Born in January 1997 in Los Alamos, he co-founded the data labeling and model evaluation company Scale AI, which was valued at $7.3 billion in 2021, with him holding about 15% of the shares. In the same year, at the age of 24, he became the youngest self-made billionaire at that time. In June 2025, Meta purchased a 49% stake in Scale AI for $14.3 billion, and Wang stepped down as CEO of Scale while remaining on the board and transitioning to Meta.
This is not a new product launch, but a restatement of the logic behind enterprise software procurement. The buyers are the procurement chains of large companies and government project leaders, while the sellers are vendors that tie deployment, narrative, and renewal together. Funds do not flow based on benchmark testing rankings, but according to who can ensure that signatories can safely endorse within the system. Benefiting are vendors that can simultaneously improve products and change narratives, with Palantir cited as a sample; under pressure are tech teams that only rely on data and screens, assuming that evaluations will automatically convert into orders.
Scale AI has previously undertaken contracts with the U.S. military and has been used by the Pentagon's Chief Digital and Artificial Intelligence Office to test the safety and reliability of large language models in military planning and decision-making. Wang has now referred to the same set of government and large client markets as a perception system rather than an evaluation system.
Source: Public Information
ABAB AI Insight
Alexandr Wang's path has always been about selling the "intermediate layer that makes models usable," not selling base models. Scale AI secured enterprise and military budgets through data labeling and evaluation, reaching a valuation of $7.3 billion in 2021; in June 2024, he launched the MEI recruitment standard, which opposes diversified hiring, turning the hiring criteria itself into an external signal. In June 2025, Meta acquired a 49% stake in Scale for $14.3 billion, bringing him along with evaluation and data capabilities, with the role of Chief AI Officer, responsible for Mark Zuckerberg's superintelligence initiatives. The controversies left by Scale are also quite specific: defense contracts, model evaluation rights, and the issue of customer isolation with Meta after his departure.
The flow of money is about buying the perception entry into the platform. Meta is not just buying labeling capacity, but placing Wang in charge of Superintelligence Labs, consolidating external government and enterprise relations, data evaluation, and internal model organization under one person. The original buyers of Scale were model companies and defense clients needing third-party evaluation and data; after the transaction, this relationship was used by Meta to bolster its model credibility narrative. The motivation is not to add an application function, but to control the narrative of "whose model is deemed usable" when open-source models and competitor labs are simultaneously scaling up.
In contrast, Palantir is not just another labeling outsourcing company. Palantir started with seed funding from the intelligence community around 2004 and has long included Keith Johnstone's improvisational theater book in its new employee reading list; Alex Karp also wrote about improvisation as an organizational method in "The Technological Republic." It sells software deployed in government and large enterprises, with renewals relying on frontline engineers rewriting internal narratives for clients. Wang is now transitioning from an independent data company owner to a platform's superintelligence leader, with an industry position closer to defense software narrative sales rather than application layer functionality competition.
Structurally, this represents a transfer of pricing power. The pricing anchor for enterprise and government orders has shifted from functional differences and benchmark scores to versions that the procurement chain is willing to publicly reiterate. The mechanism is that signatories do not bear model errors but the failure of endorsement: when reality is too unappealing, they choose a narrative that can survive within the system, with costs lower than publicly admitting system failure. Thus, improving products only addresses half of the delivery; the other half of the budget flows to vendors that can articulate deployment in the client's internal language.
ABAB News · Cognitive Laws
Evaluation wins in the lab, orders win in the narrative.
When reality is too unappealing, perception takes over the budget.
Improving products is one half; changing narratives is the other half.