The Building Blocks of Agentic Al
summary
Company-scale stack for agentic AI that targets hardware heterogeneity and on‑device inference, with concrete projects and claims: LLMs generating kernels and a Helium workflow that lets engineers author kernels in ~30 lines (claimed vs 120 in Triton) while autotuning yields equal or better performance; Monarch single‑controller orchestration for multi‑data‑center distributed training; Torch Forge as an RL/agent post‑training layer demonstrated with CoreWeave and Stanford; Exeutor 1.0 GA used at scale on devices with partners Qualcomm, ARM, Cadence, NVIDIA and AMD.
AI-written summary · about
related talks
14mArm: One Platform for Agentic AI, from Cloud to the Edge
44mBuilding Agentic AI Systems: From Transactions To Conversations
25mFrontiers of Agentic AI
30mFuture of AgenticAI: Architecting the Global Control Plane (Keynote)
39mA New Stack for the Gen AI Era
11mKeynote: Open Source Stack Empowers the AI Era
29mOpen Source Sovereign AI: Building Enterprise GenAI Platforms for Control & Comp...
29mAgentic AI: What It Is and Why It Matters