Do we need Kubernetes, or is it overkill for our team?

Kubernetes solves real problems. It also introduces an operational surface that needs someone to own it. The question is whether your problems are the ones it solves.

The threshold test

You probably need it if you are running many services with genuinely independent scaling and release cycles, you deploy multiple times a day, you need to run the same workloads across more than one environment or provider, and you have at least one person whose job includes platform operations.

If you are running a handful of services that deploy weekly on a single provider, a managed container service or even well-configured virtual machines will serve you better and cost less attention.

What it actually costs

Not the licence — the expertise. Networking, ingress, storage classes, upgrade cycles, and the debugging skill to work out why a pod is pending. Teams that adopt it without that capability end up with an outage they cannot diagnose.

The middle path

Containerise now, orchestrate later. Building images and running them under a managed service gets you the portability benefit without the control plane. When you genuinely outgrow it, the migration is straightforward because the artifacts are already right.

If you do adopt it

Use a managed control plane rather than self-hosting, standardise on one deployment pattern, and put the cluster configuration in version control from the first day. Clusters configured by hand become unrepeatable within weeks.