What we build with it
Kubernetes is where our containers actually run. We package deployments as Helm charts so that a service, its configuration, its secrets and its ingress are one versioned thing that can be reviewed, promoted between environments and rolled back as a unit.
The properties we are buying are unglamorous and specific: a rollout that halts instead of taking the service down when the new version fails its health check, horizontal scaling driven by load rather than by someone watching a graph, and a restart policy that handles the 3am crash without a human.
Where we have shipped it
Our own infrastructure, including this site, is deployed to Kubernetes via Helm. Client backends run the same way where the client's platform allows it.
We have also written about the surrounding operational work — for instance getting TLS certificates set up properly, which is part of the same ingress story.
When we recommend it
Kubernetes pays off when you run several services, need zero-downtime deploys, or have traffic that varies enough that fixed capacity is either wasteful or risky.
It does not pay off for a single small service with steady traffic. There, a managed container platform is less to operate and less to get wrong, and we will say so rather than sell you a cluster you then have to maintain.
