Deterministic Infra for Non-Deterministic AI Agents
summary
Describes a deployed agentic control plane that enforces separation: the model proposes, a policy engine approves, an execution gateway enforces, with audits and human exception handling. Shows concrete failure modes—retry amplification that turns one API error into exponential GPU consumption, memory poisoning and context corruption—and gives operational patterns: tracing of planning decisions, workload routing, elastic GPU scheduling and quotas that scale to 10k–1M runs. Details what was built and how it prevents outages.
AI-written summary · about
related talks
20mBuilding Deterministic Infrastructure for Non Deterministic AI Agents
30mFuture of AgenticAI: Architecting the Global Control Plane (Keynote)
17mAI TechOps: Data and Infra for AI Foundation to Scale and Manage 1000s of Cohesive Agents
35mThe Fight for the Future of Artificial Intelligence
27mFrom Perimeter to Silicon: Why Agentic AI Demands a New Security Model
59mArchitecting Microservices for Agentic AI Integration
22mWhen AI Agents Start Acting, APIs Become Execution Contracts
26mBeyond Sandboxes: Architecting Durable Runtimes for AI Agents