That may be true. However, one of the big advantages of using 'assistive' 'intelligent' tools over ML is explainability. If your deployment tool operates mainly on a series of constraints, it is easy to have it print a detailed log which, like the image, can be parsed through at a variable level of abstraction. So theoretically if you deployment algo is running inefficiently, you can pop it open and see why. You can edit the results before deploying. With an end-to-end ML solution it's harder to do this (at least with our current tech).

Ultimately, I think mixing many smaller solutions to mimic intelligent behavior will let us do stuff like this efficiently enough for testing.