Hand pollination is possible, of course, but that seems like such a pita. Perhaps it's possible to automate.
We have a perfectly good, self optimizing system that constantly moves to optimality. If we could just lighten up a little, not push quite so hard, or even just do localized trials of intensive use pesticides and fertilizers, we could find a balance of what the system can support.
Either go slow and look for local optimizations that are then distributed widely, or engineer immunity, or both.
Ugh. I guess if it was easy, it wouldn't be a problem.