Orchestration schedules and coordinates work: when jobs run, in what order, with which retries and dependencies. Airflow, Dagster, Prefect, and cloud schedulers do this.
Transformation changes data contents: SQL models, Spark jobs, dbt builds that turn raw tables into marts.
Airflow (orchestrate): 1) run Fivetran sync sensor 2) trigger dbt build <-- dbt (transform) 3) ping freshness checks
Common confusion
People say "Airflow transforms data" because an Airflow DAG runs SQL. Architecturally, Airflow is still the conductor; the SQL engine/dbt is the transformer.
Interview tip: "Orchestration = when/what order; transformation = how data changes." Modern stacks often pair Airflow + dbt.