Building brains, one gradient at a time