Cloud Native Mumbai: Sept Meetup

It is 2AM. Your phone is buzzing. A critical service is in CrashLoopBackOff. You are staring at kubectl output trying to connect the dots between pod events, container logs, node pressure, and resource limits. What if an AI agent could do all of that in seconds, in plain English, while you focus on the fix?

This session walks through the architecture and live demonstration of k8s-ai-agent, an open source Python-based AI agent that takes natural language prompts, translates them into Kubernetes operations, executes them safely, verifies results, and explains what it found in plain English. The session covers the agent architecture built on the LLM plus Tools plus Memory pattern, four key capabilities including pod health diagnosis, log analysis, intelligent scaling, and automated runbook generation, safe execution patterns including dry-run mode and audit logging for regulated environments, and lessons learned from building AI-driven infrastructure automation from first principles.

As a new engineer, relying on LLMs to generate Kubernetes manifests can feel like a superpower, until you run "kubectl apply" and face deprecated APIs, missing resource limits, and security vulnerabilities. In this session, we examine AI-assisted development from an early-career perspective, identifying why LLMs often hallucinate outdated configurations and create "silent failures." Rather than rejecting AI, you will learn to build a robust safety net using open-source tools like Kubeconform and Trivy to validate AI-generated code locally. Allow me to reframe your workflow: shifting from treating AI as a risky code generator to utilizing it as an interactive tutor for mastering complex concepts like RBAC, all while keeping your clusters secure and reliable.

Moving a legacy application to Kubernetes can look deceptively simple like containerize the application, create a deployment, expose a service, and let Kubernetes handle the rest. But getting an application to run on Kubernetes is very different from getting it to run reliably on Kubernetes. In my previous CNCF Mumbai talk, Breaking the Monoliths, I shared the journey of transforming a legacy enterprise application into microservices and running those workloads on Kubernetes. This talk picks up where that journey ended and explores what happens when those applications meet the operational reality of Kubernetes. Through a practical, enterprise-focused journey, we will look at the challenges that often surface after migration: legacy application assumptions, configuration and state management, health checks, networking, resource management, observability, and troubleshooting distributed workloads. We will also explore how AI-assisted operations can help engineers correlate logs, metrics, Kubernetes events, and configuration to investigate incidents faster—while keeping the engineer firmly in the loop. The goal is not to present Kubernetes as a magic solution, but to share the lessons that emerge when legacy applications move from traditional infrastructure into a cloud-native environment.