Very true, I think the issue though is unless that output is very likely to be 100% correct, a user would always prefer something that is incomplete but quicker to iterate on. It would be interesting to see if we can get to a paradigm like that.