Cloud Native AI Agents with ADK, A2A, and kagent

This talk explores the challenges of deploying and scaling AI agents in cloud-native environments, addressing critical issues like lifecycle management, security, and multi-tenancy. It introduces kagent, a Kubernetes-native runtime that enables agents to be managed as first-class workloads, leveraging familiar constructs like CRDs, GitOps, and observability, to ensure secure, scalable, and standardized agent orchestration in the modern "Agent Era."

Presented at CNCF Bangkok meet up on 28 August 2026.

Transcript

1. Cloud Native AI Agents with ADK, A2A, and kagent Fon - Kamolphan Liwprasert Data Team Lead @ AIMET.tech
2. Cloud Native AI Agents with ADK, A2A, and kagent Fon - Kamolphan Liwprasert Data Team Lead @ AIMET.tech
3. CNCF Landscape
4. CNCF Landscape - AI Native https://landscape.cncf.io/?group=ai-native
5. CNCF Landscape - AI Native https://landscape.cncf.io/?group=ai-native
6. CNCF Landscape - AI Native https://landscape.cncf.io/?group=ai-native
7. CNCF Landscape - AI Native
8. Challenges of Cloud-Native agents at Scale
9. Challenges 🔥 Sandboxing: Secure, isolated environments for code execution. Lifecycle Management: Continuous deployment, scaling, state persistence, and graceful termination. Multi-tenancy: Safely running multiple agents on shared cluster resources. Role-Based Access Control: Strict permissions for resources and APIs. Tool Connectivity: Secure and reliable connections to external databases and APIs.
10. kagent
11. kagent • https://kagent.dev : Kubernetes-native agent runtime
12. Why kagent? https://kagent.dev
13. https://github.com/kagent-dev/kagent
14. Compatibilities
15. How to deploy and operate agents?
16. Agent deployment concept AGENT ERA Microservices Era Next-Gen Agent Era Core Unit Service Core Unit Agent (CRD, Deployment, Service) Contract REST / gRPC contract Contract A2A agent card Communication Library call (In-process) Communication A2A, MCP tool call (Over network) App Framework <<various>> App Framework <<various>> Service Mesh Sidecar / Mesh Service Mesh agentgateway Gateway Ingress Gateway kgateway Topology Known at build time Topology Chosen dynamic at request time
17. Kubernetes native runtime
18. Kubernetes native runtime
19. Kubernetes native runtime
20. Kubernetes native runtime
21. Kubernetes native runtime
22. Everything as CRDs For GitOps and AgentOps! Agent CRD: Core identity, logic, and operational parameters. # Sample Agent Resource ModelConfig CRD: LLM configuration, versioning, and connection details. MCPServer CRD: Declaration of authorized remote tool servers. SandboxAgent CRD: Isolated execution environments for untrusted code. apiVersion: kagent.io/v1 kind: Agent metadata: name: sre-bot-cluster-ops spec: modelConfigRef: gpt4-prod mcpServers: - github-executor - k8s-api-server sandboxPolicy: strict
24. kagent dashboard
25. Demo
26. Agent Substrate
27. Agent Substrate: Problem statement https://www.solo.io/blog/agent-substrate-powers-kubernetes-agents-with-kagent
28. Agent Substrate https://www.solo.io/blog/agent-substrate-powers-kubernetes-agents-with-kagent
29. Agent Substrate https://www.solo.io/blog/agent-substrate-powers-kubernetes-agents-with-kagent
33. Summary
34. Thank you Fon - Kamolphan Liwprasert Data Team Lead @ AIMET.tech