Justin,
That is true but hadoop jobs are tricky so not that straightforward - some components of your jobs might not be predictable because those components also rely on data values. So considering all the aspects, yes we aim to finish you tasks within a certain time and based on that come up with the number of nodes which should be added to the cluster..Although autoscaling feature is quite comprehensive and takes care of lot of other aspects.
Would you also please describe your use case and if needed i can connect you to the team members who can give you more detailed information on your question.
thanks,
Ashish