Assets are the default, but some work is a sequence of steps rather than a set of data products: sending a batch of emails, rotating keys, calling an API to start a vendor process. For that, Dagster still has ops: @dg.op functions whose outputs feed other ops' inputs, wired together in a @dg.job (or a reusable @dg.graph, turned into a job with .to_job()).
Inside a job body you call ops like functions, send(prepare(extract())), but nothing runs then: Dagster records the wiring. Ops take config through a dg.Config parameter, keyed by the op name in run config, exactly like assets, and use resources the same way.
Ops do not produce assets by default, so they do not appear in the asset graph and have no materialization history. When the output is a table or file someone uses, model it as an asset instead. A graph-backed asset (@dg.graph_asset) combines both: several ops that together produce one asset.
Two ops wired into a job.