It's like that in other fields of engineering too, when they are making something they haven't built before. That's the essential part: for example, a lot of construction is really just rebuilding the same thing that's already been built 100,000 times in the past 100 years.

When they attempt to build something new, it often ends up like software – tremendous overruns in both cost and schedule.

C.f. the only new nuclear power plant being built in the West, which is $4 billion over budget and ten years late: https://en.wikipedia.org/wiki/Olkiluoto_Nuclear_Power_Plant#...

The reason why this feels more commonplace in software is that we're usually designing something new. Software has essentially no reproduction costs, so there's no reason for anybody to design software that's a carbon copy of something you can already download or buy off-the-shelf. That's not the case in engineering of physical products or works. New buildings are needed all the time, even if they're performing exactly the same function as the building in the adjacent lot.