NVIDIA's Omniverse Labs has unveiled nanousd-labs, an open experimental initiative that enables developers to generate lightweight, specification-compliant Universal Scene Description runtimes using AI agents. The project directly addresses a longstanding bottleneck in physical AI development: customizing implementations for constrained environments typically requires adapting expansive legacy codebases to meet specific memory, performance, or application binary interface requirements. nanousd-labs circumvents this by treating the Alliance for OpenUSD Core Specification as a machine-readable contract. AI agents systematically parse the standard, generate corresponding implementation code, and validate outputs against an automated test suite until strict compliance is verified. This methodology decouples runtime construction from the specification itself, allowing teams to regenerate optimized builds for varying workloads without sacrificing standard adherence. The resulting nanousd runtime functions as a C-based data layer responsible for parsing, composing, querying, and writing scene data, deliberately excluding rendering pipelines. Its architecture exposes a stable C ABI, enabling developers to swap backend implementations at runtime while preserving client compatibility. This design facilitates precise performance benchmarking and seamless integration with existing USD toolchains. The project originated from an internal hackathon and is now available as part of the broader Omniverse Labs open-source portfolio. Developers can engage with the initiative through two distinct pathways. Teams seeking immediate functionality can compile the existing nanousd library or utilize its Python bindings for headless, GPU-independent physical AI workflows. Alternatively, engineering groups can adopt the underlying methodology by leveraging a codified skillgraph framework. This system transforms manual agent instructions into reusable prompts and validation recipes, allowing developers to build custom specification-compliant implementations tailored to proprietary constraints. By establishing the Core Specification as a durable, machine-verifiable foundation, nanousd-labs significantly accelerates runtime development cycles and reduces reliance on monolithic distributions. The project is publicly accessible on GitHub, inviting contributions for additional language support and use cases. Alliance for OpenUSD members are additionally encouraged to participate in the Core Spec Working Group to further refine the standard governing physical AI scene description.