Self-Hosted or Managed? Architecting Secure Agentic AI Workloads

Every team building agentic AI right now is really making one decision first: build it on open-source orchestration you run yourself, or build it on a managed enterprise platform where the cloud provider owns the infrastructure. This video works through both paths properly — including the Kubernetes architecture question that comes up every time — with security placed next to every decision instead of bolted on afterward. In this video: The real first decision: self-managed vs. managed enterprise platform Path one — LangGraph, CrewAI, and Microsoft Agent Framework (the AutoGen/Semantic Kernel merger) The Kubernetes question: one cluster or two, and why separate clusters win Connecting local models to internal data: MCP, not retraining Path two — the current correct names: Bedrock AgentCore, Microsoft Foundry, Gemini Enterprise Agent Platform, OCI Enterprise AI Agents The six security controls that apply no matter which path you choose Where I'm still drawing the line between operational overhead and managed convenience Whichever path you take, six things need explicit answers and don't get lighter just because a cloud provider is managing the infrastructure underneath: identity and least privilege, data protection, network isolation, AI-specific threat monitoring, pipeline and data-source security, and governance mapped to real controls. Sources: AWS — Amazon Bedrock AgentCore Guardrails & WAF integration Microsoft Tech Community — The Great Foundry Shift Oracle — Enterprise AI Agents GA in OCI Generative AI Presenc AI — Multi-Agent Orchestration Frameworks 2026 Full write-up:https://www.linkedin.com/posts/johnka... #AgenticAI #Kubernetes #CloudSecurity #LLM #MCP #AIarchitecture #ProductSecurity #DevSecOps