Today, if you change multiple values and one of them errors, any column keys already processed are still changed however, the failure and any keys after it are not processed.
This can result in inconsistent column values that an end users may need to now reverse to retry an operation (or the app developer has to figure out how to reverse, or complete somehow).
If one column fails they should all fail, and be reverted to their pre-change values.
However I do see this may result in locks on many more columns being held until all columns are changed. (but when locks are released and changes committed only then should further automations trigger)