Couple of weeks ago I was a guest on #TheAIKubernetes Show where I had a good conversation with @wm about what it actually takes to run AI workloads on Kubernetes. Most load balancers route on URL path. The OpenAI API spec puts the model ID in the request body. Standard load balancing doesn't work for model-based LLM routing. The Gateway API Inference Extension was built specifically for this. We also got into agentic workloads and why they're different from inference, the identity and resource management challenges, and where Kubernetes is heading over the next year. Watch our conversation: lnkd.in/dE9-p-r6 #TheAIKubernetesShow #Kubernetes #AgenticAI #LLMInference