There are two types of auto scaling in Databricks: Standard and Optimized. In both scenarios when tasks are submitted the cluster will begin scaling to execute as many of them in parallel immediately.
Scaling down is different. In optimized autoscaling down will occur if the cluster is underutilized over the last 40 seconds on job clusters or 150 seconds on all-purpose clusters.
Standard autoscaling down will occur only when the cluster is completely idle and underutilized for the last 10 minutes. Standard scaling is exponential and going down will start with 1 node (scaling up starts with 8 nodes).
Check out the documentation here.