While we may not have a firm date on when AGI will emerge, one thing for certain is that 2025 will be the year of AI agents. Kubernetes is already the go-to platform for predictive analytics, MLOps, and inferencing today. What new challenges and opportunities emerge when autonomous, non-deterministic agents are deployed within our Kubernetes environments?
This talk will cover the primary building blocks of agents and multi-agent architectures and map them to Kubernetes primitives, patterns, and relevant ecosystem projects. We will discuss key challenges (and their solutions) in agentic architectures including:
- Adapting existing workloads and resources to integrate with agents via tool calling
- Observability and auditing to debug, diagnose, and explain agent behavior
- Sandboxing security context for agents to enforce principle of least privilege for the agent’s domain
- Connecting and orchestrating multi-agent interactions to support sophisticated workflows
