Complexity as the Barrier to the Next 10 Million Users

▶ Watch (0:26)

Cloud native scaled from 0 to 10 million users in under 10 years. Lin Sun heard that number at KCD Texas from Chris Aniszczyk and immediately asked the next question: how do you reach the next 10 million, faster, without adding more complexity? As CNCF TOC member, Sun surveyed community leaders and end users two years ago. The answer she kept hearing was the same: too many tools, too many things to learn, too many choices to make before writing a single line of application code.

Deploying an App Through ArgoCD With an MCP Server

▶ Watch (2:01)

Sun opened Cursor and sent a natural-language prompt asking it to create an ArgoCD application called “demo” in the default namespace using her GitHub repository. Two MCP servers were configured: an ArgoCD MCP server and a GitHub MCP server, both proxied through agent gateway. The model used was the standard Claude model, visible on screen. After an approval step inside Cursor, ArgoCD synced the repository and deployed a front end and back end, both showing healthy in the cluster within about 90 seconds.

An Agent Generating and Merging a Pull Request

▶ Watch (6:37)

With the app deployed but no HTTP route yet, Sun queried kagent, a CNCF sandbox project, to list agents running in the cluster. It returned two. She then asked the reliability agent to create an HTTP route from the agent gateway proxy to the front-end service, both in the default namespace, and to open a pull request by pulling the repository URL directly from the ArgoCD application. The agent produced the PR, the audience reviewed it live, Sun merged it, ArgoCD synced the change, and the application became reachable through the gateway.

Security Underneath: Istio Ambient Mesh and SPIFFE

▶ Watch (13:28)

Every connection in the demo, from the front end through agent gateway to ArgoCD and GitHub, ran over mutual TLS. Sun used SPIFFE for identity because the entire workload ran inside Istio service mesh in ambient mode. Ambient mode means no sidecar containers. The security properties that teams normally associate with a full sidecar mesh were present without injecting a proxy into each pod. The demo stack showed that agentic tooling and production-grade mesh security can run together without additional per-pod overhead.

What the Demo Stack Proves for Cloud Native

▶ Watch (14:51)

The four components Sun demonstrated were ArgoCD exposed as an MCP server, kagent exposed as an MCP server, agent gateway acting as an MCP gateway in front of those servers, and agent skills that let the model craft a correct HTTP route. Each piece is open source and composable. Sun’s call to action was direct: build more MCP servers, build more AI agents, build more skills, share them in the open. That stack, she argued, is the path to reaching the next generation of cloud native users.

Notable Quotes

I believe the future of cloud native is with agentic. Lin Sun · ▶ 15:16

Build more MCP server. Build more AI agents. Build more skills and let’s share it. Do it in the open and let’s keep cloud native moving. Lin Sun · ▶ 15:23

You always need a backup on stage. Lin Sun · ▶ 12:58

Key Takeaways

  • Cloud native has 10 million users; agentic AI is the proposed path to the next 10 million.
  • ArgoCD, kagent, and agent gateway can each be exposed as MCP servers and composed in a single Cursor workflow.
  • Istio ambient mesh secures all agent-to-service connections with mutual TLS and no sidecar overhead.

About the Speaker(s)

Lin Sun is Head of Open Source at Solo.io, contributing full-time to the open-source community. She serves on the CNCF Technical Oversight Committee, is a CNCF Ambassador, and maintains Istio, kgateway, and kagent. She speaks at international tech conferences.