Panels Archive

Orchestrating the Future: Heterogeneous Disaggregation at Scale for Agentic AI Workloads


Moderator: Tom St. John (Gimlet Labs)
Panelists: Zain Asgar (Gimlet Labs, Stanford University), Neeraja Yadwadkar (University of Texas at Austin), Vikram Sharma Mailthody (NVIDIA Corporation), Murali Emani (Argonne National Laboratory), Satyam Srivastava (d-Matrix)

Abstract: As agentic AI and dynamic dataflow models increasingly dominate modern compute landscapes, monolithic architectures struggle to maintain optimal cost-performance ratios. This panel explores the heterogeneous disaggregation of AI workloads, addressing the critical systems-level challenges of orchestrating complex models across diverse, next-generation accelerators.

Panelists will discuss the evolution of software infrastructure, alongside advanced scheduling strategies required to manage dynamic execution and cost-aware SLAs. By decoupling memory, compute, and networking, we can achieve highly efficient, scalable execution in fragmented environments. Join experts as they debate the future of hardware-software co-design, evaluate marginal cost-efficiency trade-offs across contemporary AI hardware, and outline the system software innovations necessary to fully realize the potential of disaggregated infrastructure at scale.


Back to Panel Archive Listing