AI Vision
Alexandr Wang's framework for where AI is headed and how it will reshape the economy, science, defense, and the global competitive landscape.
The Three Eras of AI
Wang frames the arc of modern AI as three distinct eras, each building on the last, each accelerating faster than the one before.
Pre-Training
The first era was a single, predictable curve: more compute, more data, more resources produced proportionally better models. Scaling laws were king. The roadmap was simple enough to attract the capital that built the industry.
"We were riding one very large exponential curve towards improving the performance of the models."
Alexandr Wang
Reinforcement Learning
Models learned to reason through reinforcement learning, not just predict next tokens. This was the breakthrough Ilya Sutskever saw coming before departing OpenAI. Reasoning opened the door to agentic behavior.
"What did Ilia see? Well, Ilia saw reasoning from the models."
Alexandr Wang
Recursive Self-Improvement
The most consequential era: AI systems have become instrumental in building the next generation of AI. The feedback loop is closing. Progress is no longer bounded by the number of human researchers.
"The models themselves have now become instrumental in accelerating the process of producing the next AIs."
Alexandr Wang
Personal Super Intelligence
Meta's ultimate product vision goes far beyond today's chatbots. Wang describes it as AI that becomes an extension of the self.
"Our vision is personal super intelligence. AI that knows you, your goals, your interests, and helps you with whatever you're focused on doing."
Alexandr Wang
"It won't just do your admin. It'll be an extension of you so you can be you more."
Alexandr Wang
This intelligence will be deployed across a "constellation of peripherals" (phones, glasses, wearables), always on, always contextual. Meta Ray-Bans already offer real-time translation in any language. The interaction model is wearable-first, not screen-first.
The distribution advantage is hard to overstate: 3.5 billion people use Meta platforms daily. No other AI lab can match this deployment surface.
"As we deploy powerful personal agents to everybody in the world, it creates totally new opportunities very hard for any other lab to accomplish."
Alexandr Wang
Agents Are Finally Real
Wang is refreshingly candid: "agents" was a buzzword from 2023 that "never lived up to expectations." The technology wasn't ready and the gap between demo and deployment was vast.
Something changed in late 2025. Agents started genuinely working, coding agents first, then personal agents and beyond.
"Over the course of 2026 we will see large-scale agent deployments. The GDP of AI is going to grow exponentially."
Alexandr Wang
The Progression
The role of the human shifts from doing the work, to guiding it, to orchestrating autonomous systems.
The Future of Work
"The terminal state of the economy is just large-scale humans managing agents."
Alexandr Wang
Wang draws a parallel to self-driving: it's "easy to get to 90%, hard to get to 99%." Even the best autonomous vehicles still require three to five vehicles per human tele-operator. This ratio (one human overseeing several autonomous systems) is the template for the entire economy.
"The closest thing to alchemy in our world pre-AI is programming because you can do something that creates infinite replicas."
Alexandr Wang
What was once limited to software (write once, deploy infinitely) will extend to every domain through AI agents. Every worker gains programmer-level leverage.
"The entire human workforce will soon see that large of a leverage boost."
Alexandr Wang
Wang remains optimistic about employment. His reasoning: human demand is "somewhat insatiable." As the economy becomes hyper-efficient, new categories of work will absorb the freed-up human capacity.
AI and Science
Wang believes models likely already have intuitions about biology that humans lack entirely: patterns too complex, too high-dimensional, or too subtle for human cognition to perceive.
"AIs conduct all the frontier R&D research and scientists just look at the discoveries."
Alexandr Wang
Wang sees scientific breakthroughs in biology and medicine as plausible within 12 to 24 months, a timeline that would have seemed absurd even two years ago. The roles reverse: AI conducts frontier research, human scientists interpret the discoveries.
AI and Defense
Wang describes a shift in warfare: away from "bigger bombs" and toward "fragmentation and smaller, more nimble resources." The future battlefield is populated by drones, embodied robots, and cyber warfare, not heavy armor and manned aircraft.
Scale AI built Thunder Forge with Indo-Pacific Command, compressing military decision-making cycles from 72 hours to 10 minutes. When one side operates at AI speed and the other at human speed, the advantage is insurmountable.
"Agent-driven conflict will become almost incomprehensibly fast-moving."
Alexandr Wang
China Competition
Wang's assessment of the U.S.–China AI race is more nuanced, and more sobering, than the Silicon Valley consensus.
"60-40, 70-30 that the US maintains advantage."
Alexandr Wang, on the probability the U.S. keeps its AI lead
That leaves a 30–40% chance the U.S. loses its lead. Wang's explanation for why Chinese models perform so well is blunt.
"The simplest explanation for why Chinese models are so good is espionage."
Alexandr Wang
Beyond espionage, China holds structural edges: the freedom to ignore copyright, government-run data labeling centers across seven cities, state subsidies accelerating adoption, and a manufacturing cost gap that dwarfs most estimates. A robot costing $20–30K in the United States can be built for $2–4K in Shenzhen.
2026 Inflection
"2026 is an inflection point in many ways."
Alexandr Wang
The rate of technology diffusion itself is accelerating. What took years will take months. What took months will take weeks. The compounding is relentless.
Diffusion is accelerating "two or three times faster every year." Faster development produces faster deployment, which produces faster adoption. This is what makes 2026 qualitatively different from every year before it.