It does handle schema evolution wherever it can, including inferring the initial schema automatically based on the source and destination as well, which means there's no need for manual schema changes anywhere and it will keep them in sync wherever it can.