Most teams think gpu scheduling starts with the scheduler. It starts with demand modeling. By the time Volcano, Kueue, or KEDA enters the conversation, the expensive mistake has usually already been made. The cluster was provisioned against a theoretical peak that rarely materializes. The demand curve was never drawn. The concurrency profile was assumed rather…
Your gpu utilization dashboard reads 40%. The cluster is healthy. The GPUs are loaded. Work is happening. Except it isn’t. That 40% gpu utilization figure is a peak average across a monitoring window. What it doesn’t show is the seven minutes before that spike when every GPU in the cluster was resident in memory, warm,…
The LLM security boundary problem isn’t a Kubernetes misconfiguration. It’s a category error. You’re applying infrastructure isolation to a system whose failure mode is behavioral. Kubernetes was designed to answer one question: is the workload running correctly? It answers that question well. But when the workload is a large language model, “running correctly” and “behaving…
The CLI control plane is the most powerful — and least governed — layer in most enterprise infrastructure stacks. Most infrastructure changes don’t happen in dashboards. They happen in terminals. The UI is where you observe. The CLI is where you commit. A console shows you a representation of state. A CLI invocation changes it…
Agentic AI control plane governance is the architecture problem most teams are not modeling — and the one that will produce the most expensive failures in 2026. The control plane became the most sensitive layer in modern infrastructure. So we locked it down. Kubernetes gave us control plane isolation — the API server, etcd, and…
The control plane shift is the most important infrastructure concept of 2026 — and most teams are experiencing it three or four times simultaneously without recognizing it as the same decision each time. Your VMware renewal lands on the desk. The number is larger than last year. You open a spreadsheet and start modeling Nutanix….
I’ve watched observability vs monitoring play out as a live incident more times than I can count. The dashboard was green. The on-call engineer was not paged. The monitoring system did exactly what it was designed to do — it watched for thresholds, waited for metrics to cross them, and stayed silent when they didn’t….
The pitch for AI in engineering was straightforward: automate the repetitive, accelerate the cognitive, and let engineers focus on higher-order problems. Less time writing boilerplate. Less time provisioning infrastructure. Faster feedback loops. Lower operational overhead. Some of that happened. But something else happened too — something nobody put in the pitch deck. The AI systems…
>_ AI Inference Cost — Series Part 1 — Cost Architecture AI Inference Is the New Egress: The Cost Layer Nobody Modeled Part 2 — Execution Budgets Your AI System Doesn’t Have a Cost Problem. It Has No Runtime Limits. Part 3 — Model Routing Cost-Aware Model Routing in Production: Why Every Request Shouldn’t Hit…
