I think the analogy (and it is not to be taken literally) is that of "commoditized processes".
Nowadays we don't build bridges to suit the site, we choose sites to accommodate bridges that we basically build identically via a few designs.
Connecting back to s/w AI can do the standard stuff ok as long as you test around the outside of it, so you might want to hone your judgment about how you build systems so it uses the stuff AI can do well, vs "building for the site". The gains are productivity. The losses are efficiency (the problem must go through some extra steps to meet the process where it works). Same as any engineering problem at scale.
I think the analogy (and it is not to be taken literally) is that of "commoditized processes".
Nowadays we don't build bridges to suit the site, we choose sites to accommodate bridges that we basically build identically via a few designs.
Connecting back to s/w AI can do the standard stuff ok as long as you test around the outside of it, so you might want to hone your judgment about how you build systems so it uses the stuff AI can do well, vs "building for the site". The gains are productivity. The losses are efficiency (the problem must go through some extra steps to meet the process where it works). Same as any engineering problem at scale.