Alibaba Cloud's Peng Li Discusses Scaling Stateful AI Agents to Zero with Knative
A video published on the PyTorch official channel features a talk by Peng Li from Alibaba Cloud. The presentation explores how to combine typically always-on stateful AI agents with Knative's elastic, scale-to-zero services by externalizing agent state, separating session continuity from execution, and pooling heavy runtimes to achieve serverless elasticity.
ImportanceLocalEvidenceE2 unreplicatedWrite-upQuick
In a video published on the PyTorch official channel on September 24, 2026, Peng Li from Alibaba Cloud delivered a technical talk titled "Can Stateful AI Agents Scale To Zero?" This resource is relevant for engineering teams focused on AI infrastructure cost optimization and architecture design.
AI agents are often treated as always-on workloads because they maintain session context, invoke tools, and may depend on sandboxed runtimes. This creates a natural conflict with elastic, scale-to-zero service architectures like Knative. In his talk, Peng Li demonstrated how these two concepts can work together: by externalizing agent state, separating session continuity from execution, and pooling heavy runtimes, stateful agents can still benefit from serverless elasticity. The presentation also covered the boundaries of this model, discussing scenarios where it works well and where always-on runtimes remain the better choice.