On-Demand: Getting Started with Agent Substrate: Run AI Agents at Scale with kagent
AI agents are long-lived, stateful, and idle much of the time, which makes running them like traditional Kubernetes workloads inefficient. Agent Substrate introduces a new runtime designed specifically for agentic workloads, while kagent provides the Kubernetes-native framework for building and managing those agents. Together, they enable secure, sandboxed execution, fast startup times, and significantly better infrastructure utilization.
In this hands-on workshop, you'll learn how Agent Substrate works, why it exists, and how to deploy your first sandboxed agent with kagent.
You'll learn how to:
- Understand the Agent Substrate architecture and runtime model
- Deploy kagent with Agent Substrate on Kubernetes
- Create and run sandboxed AI agents
- Explore WorkerPools, Actors, and agent lifecycle management
- See how agents are scheduled, isolated, and managed at runtime
You'll leave with:
- A working kagent + Agent Substrate environment
- A solid understanding of when and why to use Agent Substrate
- Practical experience deploying sandboxed agents on Kubernetes
Frequently asked questions
What is Agent Substrate?
Agent Substrate is a runtime designed for agentic workloads on Kubernetes. AI agents are long-lived, stateful, and idle much of the time, so Agent Substrate runs them as sandboxed Actors in WorkerPools, with fast startup and efficient handling of idle agents instead of dedicating a full pod to each one.
How does kagent use Agent Substrate?
kagent is the Kubernetes-native framework for defining and managing AI agents. Deployed with Agent Substrate, kagent runs agents as isolated, sandboxed Actors, so you keep kagent's declarative agent definitions while gaining secure execution, fast startup, and better infrastructure utilization.
Why are AI agents inefficient on stock Kubernetes?
Standard Kubernetes workloads assume services that are busy most of the time. AI agents often sit idle between tasks while still holding a pod's reserved resources, and they need strong isolation for the code and tools they run. Running each agent as a regular pod wastes capacity and slows startup; a runtime built for agents addresses both.
Speakers
