VCs Eye 'World Models' as Next AI Frontier

The concept of "world models" traces back to a 2018 paper by David Ha and Jürgen Schmidhuber, which proposed that an AI could learn a compressed, internal model of its environment. This allows an agent to "dream" or simulate future scenarios within this learned model, drastically improving learning efficiency for tasks like controlling a race car in a game. This is a fundamental departure from LLMs, which predict the next word, to systems that predict the consequences of actions in a simulated reality. Meta's Chief AI Scientist, Yann LeCun, is a major proponent of this approach, arguing that world models are a crucial step toward human-level AI because they enable reasoning, memory, and planning. He envisions embodied AI systems that learn about the physical world through interaction, a vision he's pursuing after leaving Meta to found a new startup focused on Advanced Machine Intelligence. This move signals a broader industry shift toward AI that can understand and interact with dynamic, real-world environments. In NYC, this frontier is already being explored. Companies like RunwayML are explicitly building AI to "simulate the world" and are hiring for roles in Physical AI to turn research into customer outcomes in robotics and autonomy. General Intuition, a frontier research lab in NYC, is hiring Applied AI Engineers for "Spatial and Embodied AI" to build foundation models for environments requiring spatial and temporal reasoning. Additionally, robotics startup Reflex Robotics is hiring a Principal AI Research Engineer for World Models in NYC to help build the largest robotics dataset in the world. For engineers looking to build, the NYC startup scene offers tangible opportunities. IronLedger.ai (a YC S25 company) is hiring founding engineers in New York to build AI agents for property accounting. Open-source hubs like Hugging Face also have a significant presence in NYC and are hiring for roles like "Agent Product Manager" and engineers for "Agentic AI," indicating a focus on this next wave. VCs in NYC are taking note. FoundersX Ventures recently opened a new office at 3 World Trade Center and is actively investing in robotics, world modeling, and embodied AI. Karman Ventures, another early-stage firm, explicitly lists "Embodied AI" as a core area of interest for its pre-seed and seed investments. This local investor interest provides a potential funding path for founders building in this space. Engineers can start building practical AI agents on the side today. One indie hacker created an agent to handle annoying customer service calls, combining planning, email, and phone call tools. Another built and sold an AI agent for a travel company that reads PDFs and Excel files to find the best rates. These projects often start with frameworks like LangChain or even no-code tools like N8N to automate real-world workflows, providing a blueprint for turning a side project into a potential business. For those interested in consumer apps, Gen Z's embrace of AR and virtual try-on technology offers a glimpse into the demand for simulated experiences. Over 90% of Gen Z shoppers are interested in using AR, not just for convenience, but for self-expression and creating digital identities. This behavior points to opportunities in building social and e-commerce apps where users can interact with products and each other in rich, 3D-simulated environments. The user's background in insurance also presents vertical SaaS opportunities. Property and casualty insurers use simulation for dynamic financial analysis and to model the impact of catastrophic events. An engineer could build an AI agent that runs "what-if" scenarios for risk management or uses predictive modeling to identify high-risk claims, offering a targeted solution to a high-value industry problem.