英伟达：2026年AI编程赋能GPU开发者：构建编码智能体实现“光速代码”开发报告（英文版）

We commit to • Build the reference implementation & get out of your way • Productize what needs NVIDIA scale & expertise Delivering Intelligence • Full stack FOSS AI v RAG • -> MCP (Model Context Protocol) • -> SKILLs • -> Agents select Skillsets, intrinsic evals Agentic Coding - Performance RL• Balance Speed & Memory Pressure• Also token cost, time• Maintainability - CUDA-X, PTX Fragile