What would you include in an Airflow DAG, and what would you avoid adding?

In an Airflow DAG you should include the core elements that define a reproducible, maintainable workflow: a unique dag_id, a sensible schedule_interval, start_date, and catchup flag; default_args that set retries, retry_delay, email_on_failure, and owner; a clear list of tasks (operators) with explicit dependencies (set_upstream/set_downstream or bitshift operators); tags and documentation for discoverability; and optional sensors or triggers for event‑driven execution. You should also add logging, XComs for lightweight data passing, and a clear error‑handling strategy. What to avoid: embedding heavy business logic inside a PythonOperator; hard‑coding credentials or secrets in the DAG file; using long‑running tasks that block the scheduler; creating too many parallel tasks that exceed the cluster’s capacity; using global variables that break idempotence; and adding non‑deterministic or non‑idempotent operators. Keeping the DAG declarative, lightweight, and idempotent ensures it remains maintainable and scalable.