So, even when we do roughly know how a optimal design would look like, we often have to do incremental development and I think finding ways to make it easier is a valid problem at least, even if the biological metaphors aren't directly applicable.
However, I've never found a way to make it actually work in a real system. The cost of redundancy is high, and you absolutely need a way to pass around the chain of configuration choices with the values. Otherwise, if something goes wrong, or even to understand the values you're getting, you're stuck grubbing through log files (if you're lucky) to find out what happened.
I suppose his propagator research is a way to make this easier to do, but it's so specialized that it becomes an all-or-nothing approach, not something you can incrementally develop towards.
