At Siggraph this week, Nvidia outlined a vision for hybrid AI in computer graphics, where AI agents orchestrate established graphics and physics tools. Nvidia is extending its modular, agent-friendly philosophy to Omniverse by expanding its Agent Toolkit with new Omniverse libraries—software components that give AI agents tools to add physical AI capabilities and prepare 3D content for simulation. These libraries expose core capabilities like RTX rendering and physics simulation in an ‘agent friendly’ form. Nvidia showcased integrations including PTC Onshape for CAD-to-sim-ready asset generation. A key theme is the limitation of today’s neural graphics and generative models with fine-grained control; Nvidia’s hybrid approach uses classical techniques inside AI-driven workflows. Nvidia presented 21 papers of research in this area, including ArtiFixer, an AI model that turns messy real-world 3D captures into clean, complete virtual scenes, predicting photoreal global illumination from geometry without tracing a single ray. Nvidia expects major uptake in architecture, city planning, GIS, and training robots. To make AI agents practical, Nvidia is promoting Model Context Protocol (MCP) for agents to work with existing DCC tools.