Sandboxed Agents: Building Isolated, Governed and Scalable AI

How do you give an AI agent the tools it needs while controlling which files, environments, models, and services it can access? And how do you run many agents efficiently when their activity rises and falls? Join the inaugural AAIF Boston event, sponsored by solo.io for a technical session on Agent Substrate, an open-source runtime for managing sandboxed agent workloads on Kubernetes. We will explore how Kubernetes primitives and an agent-specific control plane work together, how Substrate Actors share a pool of workers, and what that means for isolation, resource usage, and responsiveness. The session will examine the idea of running 200+ agents across a small pool of Kubernetes pods, including the workload assumptions behind that density. What you will learn - How Kubernetes primitives support the infrastructure for sandboxed agents. - How an agent-specific control and management plane handles actor placement and lifecycle. - Where to enforce access to files, commands, production environments, model endpoints, MCP tools, and credentials. - How activity patterns, suspend/resume behavior, and worker capacity affect scaling and performance. Speaker - Michael Levan, Principal AI Architect/FDE Who should attend AI engineers, platform engineers, SREs, security engineers, FDEs, and developers building or operating agent systems. Familiarity with Kubernetes will help. This is a technical talk and discussion. About AAIF BostonAAIF Boston brings together developers, researchers, and practitioners to share technical work, useful artifacts, and lessons from building agentic AI. Events are vendor-neutral and focus on engineering practice. All participants must follow the Linux Foundation Events Code of Conduct. Registration information is handled under the Linux Foundation Privacy Policy. Learn more about Agent Substrate ---