What job interviews taught me about Kubernetes

That framing mostly held up. People who have lived through VM plus systemd, hand-built EC2 fleets, and bespoke deployment setups said managed Kubernetes now feels more boring than those older stacks, especially once you care about repeatability, rollback, drift correction, and replacing the one infra person who knows all the weird parts. Several commenters sharpened the point: a lot of what the post praises is really infrastructure as code and standardization, not something unique to Kubernetes. Kubernetes still wins because the abstractions are more uniform across clouds than Terraform stacks or provider-native services, and because the cluster keeps enforcing desired state after deploy rather than just provisioning resources once. The pushback was blunt. People who regretted early Kubernetes adoption said newcomers underestimate how little comes in the box. A “basic” setup quickly turns into ingress, certificate management, DNS controllers, storage integrations, networking plugins, and a steady stream of version churn and breaking changes. The burden is not only initial setup. It is owning the operational surface area of the controllers around the cluster and keeping up with the upgrade treadmill. That is why several people argued the better comparison for a startup is not “Kubernetes or chaos” but “Kubernetes or a much smaller deployment layer” like Podman plus Ansible, Nomad, ECS, Cloud Run, Kamal, or simply cloud-native infrastructure tooling. The fault line was not really pro versus anti Kubernetes. It was whether recent tooling changes have shifted the cost curve. One camp said managed offerings like EKS and lighter distributions like k3s have made startup-sized clusters routine. They also said LLMs are unusually good at generating manifests, Helm values, Terraform, and dashboards, which cuts a lot of the old yak shaving. The other camp said that makes setup easier, not operations safer. The hard parts show up later in networking, cloud integrations, egress policy, upgrades, and debugging weird failure modes. The useful conclusion is narrower than the post’s title suggests: Kubernetes has become the default organizational interface for running software, but that does not mean it is the simplest operational choice for every small team.