For much of the past decade, companies moved data into their analytics systems with pipelines written as code: Spark or Python jobs. The approach works, but it is brittle. A single change upstream, ...