A new approach to building AI agents is emerging from an unexpected source: video game footage. General Intuition, a frontier AI research lab, is leveraging billions of action-labeled gameplay clips to train the next generation of agentic systems that can perceive, predict, and act in complex environments. The company raised $320 million in Series A funding at a $2.3 billion valuation in 2026, signaling strong investor confidence in this unconventional path to embodied artificial intelligence.

Traditional AI agents have relied heavily on language models and text-based training data. But General Intuition is taking a different route by tapping into Medal.tv, its sister platform, which hosts billions of action-labeled gameplay videos. This unique dataset provides something that standard text and video datasets cannot easily deliver: rich spatial-temporal reasoning data that shows how agents should perceive, predict, and act in dynamic environments.

What Makes Gameplay Data Different for Training AI Agents?

Gameplay videos are fundamentally different from other training data because they capture real-time decision-making, cause-and-effect relationships, and physical interactions. When a player navigates a game world, jumps over obstacles, or solves environmental puzzles, the video records not just what happened, but why it happened. This "action-labeled" aspect means researchers can teach AI systems to understand physics, causality, and intelligent decision-making in ways that pure language training cannot achieve.

The implications extend far beyond gaming. General Intuition is building what it calls "world models" and "action models," systems that develop what the company describes as a true "general intuition of reality." These models are designed to improvise, predict outcomes, and act intelligently across diverse environments, from virtual simulations to physical robotics applications.

How to Understand General Intuition's Approach to Agentic AI

• World Models: Systems trained to understand physics, causality, and how environments respond to actions, building on prior research like IRIS, DIAMOND, and GAIA-2.
• Action Models: AI components designed to make intelligent decisions and predict outcomes based on environmental understanding, enabling agents to act effectively in unfamiliar situations.
• Cross-Domain Applications: The technology targets gaming (better non-player characters and agents), robotics, autonomous systems, and enterprise simulation, showing versatility beyond any single use case.

The $320 million Series A funding reflects investor appetite for approaches that move beyond the current limitations of language-only AI agents. While large language models (LLMs) excel at text generation and reasoning tasks, they struggle with embodied understanding, the ability to grasp how physical actions produce real-world consequences. General Intuition's gameplay-based training methodology addresses this gap directly.

The company's roadmap is ambitious. It plans to rapidly scale compute resources and model capabilities while recruiting top-tier talent and forming strategic partnerships. However, the lab remains in early stages, with commercial API access rolling out gradually. Production results in real-world environments are still emerging, and the approach demands significant computational resources typical of frontier model development.

Why Does This Matter for the Future of AI Agents?

The agentic AI field has been dominated by discussions about speed, security, and tool use. But General Intuition's approach highlights a different challenge: how to build agents that truly understand the world they operate in. By training on billions of gameplay interactions, the company is attempting to create AI systems that don't just follow instructions or use tools, but develop genuine intuition about cause and effect.

This shift has practical consequences. Better world models could lead to more capable autonomous systems in robotics, more realistic and adaptive non-player characters in games, and enterprise simulation tools that behave more like the real systems they're meant to model. For enterprises exploring advanced agentic AI, this represents a fundamentally different approach than fine-tuning language models or adding function-calling capabilities.

The research pedigree behind General Intuition adds credibility to its ambitions. The company has published breakthroughs in embodied AI and world modeling, and its backing from major investors suggests the approach has technical merit beyond the hype cycle that often surrounds AI startups. As agentic systems become more central to enterprise operations, the ability to build agents that understand physical and logical causality may become a competitive advantage.