Building World Models for the AI-Native Economy

A frontier AI research lab for understanding what comes next

The economy is one of the hardest environments for AI to understand. It is only partially observable, constantly changing, and shaped by interactions between technology, companies, consumers, and markets. Predicting how this system evolves may seem close to impossible. We believe it is one of the most consequential research problems in AI.

ViserAI is building systems that learn an internal model of the economy. Our models aim to discover the latent state of companies and markets, reason about how that state may change, and revise their predictions as new evidence arrives. The long-term goal is to identify important risks and opportunities earlier and help direct capital toward technologies and companies that drive real economic progress.

Our Research Direction

01

A world model for the economyMarkets are not static datasets; they are living systems. We are building AI that learns how companies, industries, and technologies evolve by forming an abstract, continuously updated representation of the economy.

02

Learning from experienceThe system is designed to learn from change rather than depend on fixed labels or directly predict prices. It observes signals, builds an understanding of what matters, and improves that understanding as the world reveals what happened.

About the Team

ViserAI was founded by researchers and engineers who previously built and led AI systems at Google DeepMind, Cohere, and Verily. We have taken ambitious ideas from first principles to influential research and production-scale systems.

We helped create major advances in self-supervised learning and predictive representations (BYOL, PBL), world models and exploration (World Discovery Models, Noisy Networks, BYOL-Explore), and language-model alignment and self-improvement (IPO, SRPO).

The scientific direction is now receiving major validation. AMI Labs, Yann LeCun’s world-model company valued at $3.5 billion, is pursuing a closely related principle for the physical world: learning abstract representations and predicting in representation space. Our earlier work on BYOL and PBL is part of the research lineage behind this broader direction. At ViserAI, we are extending the same big idea to companies, industries, and markets.

We are now applying this research lineage to the economy and building a small, exceptional team across reinforcement learning, representation learning, time-series models, causal inference, and research engineering.