The Reasoning Show Kagenti - A Kubernetes Control Plane for AI Agents
32 snips
Mar 18, 2026 Morgan Foster, an engineer focused on distributed systems and AI in Kubernetes, co-chair of the Kubernetes AI Gateway WG and maintainer on Kagenti. He describes Kagenti as a framework-neutral control plane for agentic AI, talks identity and zero-trust for agents, orchestration vs data-plane tradeoffs, and challenges of multi-tenant, unpredictable agent behavior and safe tool access.
AI Snips
Chapters
Transcript
Episode notes
Kagenti Provides Framework Neutral Agent Middleware
- Kagenti is an agent-framework-neutral middleware for Kubernetes that extracts common platform needs like discovery, identity, and workload management.
- Morgan Foster merged prototypes from IBM Research and his own work to provide discovery via A2A primitives and workload identity federation.
Agents Are Unpredictable Systems Requiring New Observability
- Agents differ from services because their behavior is unpredictable and probabilistic, requiring new observability and profiling approaches.
- Morgan Foster argues we must build profiles and richer traces to detect drift rather than rely on static health indicators.
Use Federated Short Lived Identities For Agents
- Give agents a robust, short-lived federated identity rather than just a cluster-scoped service account.
- Kagenti issues SPIFFE IDs and federates SPIRE with external identity providers to enable quick onboarding and teardown.