Treat your cluster like the batch scheduler it already is

What it does Argo Workflows is a Kubernetes CRD that orchestrates parallel jobs by treating every workflow step as a container. You define dependencies as a directed acyclic graph or a simple sequence, then let the cluster handle scheduling, retries, and artifact routing between steps. It targets compute-heavy work like machine-learning pipelines, data processing, and CI/CD.

The interesting bit The clever part is that workflows are native Kubernetes custom resources, so steps run as ordinary pods and reuse the cluster’s own scheduler, node affinity, and garbage collector rather than hiding them behind a separate abstraction.

Key highlights

• Steps are containers; artifacts pass between them via S3, GCS, Azure Blob, Git, HTTP, and more
• Declares workflows as DAGs or sequential steps with loops, conditionals, parameterization, and cron scheduling
• Bundles a web UI, REST and gRPC APIs, a CLI, and SDKs for Java, Go, Python (Hera), and TypeScript (Juno)
• Supports SSO via OAuth2/OIDC, Prometheus metrics, webhook triggers, and Pod Disruption Budgets
• CNCF graduated project with roughly 200+ listed organizational users and an ecosystem spanning Kubeflow, Metaflow, and Argo Events

Caveats

• The README bills it as the “most popular workflow execution engine for Kubernetes,” but supplies no citation or download data to support that claim
• With dozens of features from DinD to Windows container support to multiplex log viewing, the surface area is broad enough that “light-weight” depends heavily on which controllers and executors you actually enable

Verdict Evaluate it if you already run Kubernetes and need to schedule batch pipelines, ML training, or infrastructure automation without maintaining a second orchestrator. Skip it if you are not on Kubernetes or your workflows are simple enough for a few shell scripts.