Prefect

Prefect positions itself as the Pythonic alternative to Airflow: define workflows using decorators on regular Python functions, run them locally or at scale, and observe everything through the Prefect Cloud or self-hosted UI. The framework has been through a major architectural rewrite (Prefect 2 in 2022, now Prefect 3), which broke compatibility with the original 1.x API and produced the documentation instability that shows up clearly in practitioner feedback.

For teams already comfortable with Python, Prefect's task and flow model is natural — pipelines feel like Python, not like DAG configuration. The hybrid execution model (Prefect Cloud controls orchestration; user infrastructure runs the actual tasks) lets teams keep data and compute on-premises while offloading scheduling and observability to the managed cloud. Self-hosted Prefect Server is available for teams that cannot use external control planes.

The documentation friction is real. The 1→2 migration broke substantial community resources, and the Prefect 2→3 release introduced further changes. Practitioners evaluating Prefect frequently encounter outdated examples or tutorials written for the wrong version. Kubernetes-based deployment for production requires infrastructure setup that takes more time than cloud-managed alternatives. Concurrency APIs have been noted as less mature than Airflow's for complex parallel workflows.