Auto Mode is managed Karpenter plus managed node lifecycle plus managed core add-ons.

You describe workloads; AWS provisions, patches and replaces the nodes. You trade fine control for almost no node operations.

Standard EKS (you run)

Node groups or self-run Karpenter and its NodePools
Node AMI selection and update rollouts
Install and update CNI, CoreDNS, kube-proxy, CSI, load balancer controller
Upgrade runbooks per layer

Auto Mode (AWS runs)

AWS provisions nodes from Pod demand with managed Karpenter
AWS-managed node image, patching and replacement (nodes have a maximum lifetime)
Built-in networking, storage and load balancing capabilities
Default node pools you can tune; AWS controls the controller version

Choose it when

You want Kubernetes' API without owning node operations and are fine with AWS's defaults and cadence.

Think twice when

You need custom AMIs, SSH or node-level agents with special needs, or very fine control over instance selection.

Cost and rotation

An extra management charge applies per node on top of EC2. Nodes are recycled on a schedule, so workloads must tolerate rotation.

enable on a new cluster

$ eksctl create cluster --name shop-eu --enable-auto-mode

EKS Auto Mode: AWS operates the nodes, and more

EKS Auto Mode extends the managed-compute idea. Instead of you installing and running Karpenter and the core add-ons, AWS runs them: it provisions and scales nodes in response to Pod demand, manages the node image and patching, replaces nodes on a schedule (nodes have a maximum lifetime, so workloads must tolerate regular rotation), and operates networking, storage and load-balancing capabilities as part of the service. Your interaction is mostly at the Pod level.

The trade-off is control. Compared with self-run Karpenter you lose custom AMIs, direct node access and fine-grained control of the controller. In return you eliminate a large amount of node-layer toil, including upgrade runbooks for the data plane. There is a per-node management charge on top of EC2 pricing, and it does not shrink with Savings Plans, so compare the total cost of ownership including the engineering time you save.

A simple decision rule: if your platform team wants to avoid node operations and your workloads do not depend on node customisation, Auto Mode is a strong default for new clusters. If you need custom AMIs for compliance, specialised DaemonSets, or tightly tuned instance selection, keep standard EKS with Karpenter or managed node groups.

These details are evolving quickly. Features, pricing and limits of Auto Mode change; confirm in the current documentation, and test with representative workloads before migrating a production cluster.