From Chat to Action: Why General Intuition’s $300M Raise Signals a New Era of Agentic AI

AI is shifting away from models that simply "talk" and toward systems that can "act." General Intuition, a New York-based powerhouse in the spatial reasoning space, is currently at the forefront of this transition, seeking to secure approximately $300 million in funding. This funding shows investors are now looking for infrastructure that supports "Agentic AI," where machines are not just predicting the next word in a sentence but are navigating and interacting with complex, three-dimensional environments.

Current Large Language Models (LLMs) have limits. While they are great at language, they lack an inherent understanding of physical causality—the logic of how gravity works, how objects occupy space, and how physical actions lead to specific consequences. To bridge this gap, General Intuition focuses on "World Models." These are AI systems designed to understand the laws of physics by training in simulated environments. By creating a system that understands the "how" and "why" of moving through a 3D space, the company is laying the groundwork for the next generation of autonomous robotics and industrial automation.

Why is video game data the "secret sauce" for spatial intelligence?

General Intuition’s core innovation is its choice of training ground: billions of video game clips. While this might sound like a simple gaming play, it is actually a sophisticated strategy to overcome the scarcity of high-quality, structured physical data. Unlike standard internet videos or text, video games are governed by "physics engines." In a game world, every interaction follows consistent rules; if a character pushes an object, that object moves according to predefined laws of weight and friction.

By training on these clips, General Intuition’s models develop what is known as "spatial intelligence." They aren't just watching pixels change; they are learning the underlying logic of the environment. This gives them a clear advantage over regular video data because it provides consistent feedback, teaching agents how to move around and handle objects.