Kubernetes is not a rite of passage. It is a response to concrete pain: multiple services, health-based rollout, horizontal scale, and environment parity.
Growing teams get value from managed control planes, Helm (or equivalent) packaging, and a small set of golden paths — not from inventing a unique snowflake cluster per product.
If you cannot yet observe a Compose stack, Kubernetes will not save you; it will multiply the unknowns.
Adopt when the pains are real
Signals: several deployable services, need for rolling updates, autoscaling, and clearer separation between app and node lifecycle.
Counter-signals: one app, one database, one team, rare releases — start simpler.
Golden paths over infinite flexibility
Provide a paved road: build → image → chart → deploy → dashboards. Optional complexity should be optional.
Standardize probes, resources, and secrets early.
Cost the platform
Include cognitive load. A platform that only the original author can debug is not a platform.
Takeaways
- Orchestrate for pain you have, not status you want.
- Paved roads beat policy PDFs.
- Observability first — then scale out.