We use them for things that require unpredictable but immediate bursts of scaled parallel processing