Really like their concept of assets and compute.
Defining the desired results of your pipeline, the transforms that create those assets, and then letting Dagster build/materialize those results.
I’ve spent a lot of time in other tools restarting jobs, asking questions like “did that job run?” and “is this using the latest stats output?”
Dagster’s UI of “this data pipeline is current as of 12:30” gives the answers our product team needs without pain.
And for the same reason the branch compute makes a lot of sense. “The output with the updated algorithm is Y” makes a lot more sense than having a separate experiment pipeline
