Kubernetes

advancedwidespread

An open-source system that automatically starts, monitors, and restarts programs packaged in containers on many servers, and adjusts their number to the load depending on the settings.

In plain terms

Today's online services rarely run on a single machine. They consist of many smaller program parts packaged into so-called containers: these are uniform boxes containing the program and everything needed to run it. Kubernetes decides which box goes to which server and ensures that enough instances are always running.

Imagine the control tower of a large container port. The ships are servers, the containers are programs. The controller continuously monitors which ship has free space and what has broken down. If a ship drops out, its cargo is reloaded onto another ship, so shipping continues without interruption.

In the AI world, this is important because serving large language models and AI agents is expensive, takes place on graphics card-equipped machines, and the load changes hour by hour. Kubernetes allocates these machines. When properly configured, the service remains available during peak times and uses fewer machines during quiet periods.

How it works

Developers describe the desired state in a configuration file: for example, that five instances of the shopping cart service always run in the online store.
The control plane checks how much free processor, memory, or graphics card capacity is available on each server and schedules the containers accordingly.
The system continuously compares reality with the desired state: if a container crashes or a machine fails, it starts a new one elsewhere without being asked.
It distributes incoming requests among running instances. If automatic scaling is enabled, it launches more instances under high traffic and shuts down excess ones under low traffic.
During an update, the old version is gradually replaced by the new one, so users ideally notice nothing, and in case of an error, it is possible to revert to the previous version.

An everyday example

At 8 p.m., several hundred thousand people start asking an AI chatbot questions simultaneously, yet you still receive an answer within a few seconds. In the background, a Kubernetes cluster likely launched additional instances of the model, and by dawn, when traffic is low, it shuts them down.

Who is affected

developerMany modern applications and AI services run on Kubernetes, so its fundamentals offer an advantage in many developer roles or are explicitly required. → Code

entrepreneur, executiveIt greatly influences how reliable and how costly your company's online service is, and how independent you remain from a single cloud provider.

jobseeker, new graduateKubernetes knowledge is one of the most sought-after skills for operations, DevOps, and AI infrastructure roles. → Work & HR

everyoneOnline banking, streaming, e-commerce, and the AI chatbot you use daily often run on such systems in the background.

A common misconception

MythKubernetes is a cloud service that must be purchased from Google.

In realityIt is free, open-source software maintained by the non-profit Cloud Native Computing Foundation, which belongs to the Linux Foundation. You can run it on your own machines or on any major cloud provider. You pay for the machines and the managed convenience services.

Limits and risks

Complex system: its setup and operation require significant expertise, often an unnecessary overkill for a small website or a team of a few people.
It does not make the application secure by itself: poorly configured permissions and left-open interfaces are common attack points.
Due to poorly specified resource requirements and unnecessarily running instances, the cloud bill can silently skyrocket, especially with expensive GPU-equipped machines.
It only organizes the execution: it does not fix a poorly written program or a weak AI model.

Similar, but not the same

Docker – Docker is primarily used for creating containers and running them on a single machine. Kubernetes, by contrast, orchestrates the coordinated operation of many containers across many machines.
virtual machine – A virtual machine emulates a complete computer with its own operating system, while Kubernetes manages lighter containers, which often run on virtual machines themselves.
Cloud service – The cloud provides rental machines and storage, while Kubernetes is the software layer that organizes how programs run on those machines.

Related concepts

Broader

container orchestration, Cloud computing, AI infrastructure

Narrower

Pod, Cluster, automatic scaling, Helm

On the edge

Container, Docker, Microservice, DevOps, MLOps

A brief history

Google released it as open source in 2014, building on its experience with its own internal systems, primarily Borg. Version 1.0 came out in July 2015. That was when the Cloud Native Computing Foundation was formed, and Google handed the project over to it. The name means helmsman in Greek, and the abbreviation K8s refers to the eight letters between the K and the s.

Where things stand

As of late 2026, it is the de facto standard for operating containerized applications, and every major cloud provider offers it in managed form. It is also one of the most common foundations for training and serving AI models and for running agents.

Status: 2026 November

In the news

Appears in 1 Agent Age stories; the curve shows weekly occurrence in the news magazine.

Test of six AI coding agents on DevOps tasks: Terraform, Kubernetes, Ansible, DockerSep 22, 2026 · Code

Also known as

Kubernetes · K8s · k8s · Kube · Kubernetes cluster · Kubernetes-klaszter · Kubernetes cluster

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.