Hi @Michal Mlaka​ , it should be possible. If you are using APIs to launch the Job, remove  "node_type_id" and "driver_node_type_id" from the JSON and pass the "instance_pool_id" and "driver_instance_pool_id" instead to make the cluster pick VMs from the pool.

Check this doc for the structure of cluster definition-

https://learn.microsoft.com/en-gb/azure/databricks/dev-tools/api/latest/clusters#--request-structure...

Let me know if this helps. Please mark the comment as "best answer" to resolve the query.