I’ve thought a lot about this and experimented with versions of it but I just can’t see it being worthwhile. The cost to undo any decision made at this level is swiftly trending to zero
Soon, your agents will be able to trace code back to the threads that produced it, giving them access to the decisions and reasoning the code alone can’t explain.
