We desire to make our creations simultaneously as capable and as flexible as possible, but that only means making them more and more complex. And with increasing complexity comes decreasing robustness. The two properties are antithetical, and at cross-purposes to one another. Good engineering can optimize systems to come up with a resonance of redundancy and performance, and inspired engineering can give us systems that are modifiable and expandable, but sooner or later it bites us in the ass. Every problem or error we find and correct in an old design only adds another unanticipated and untested fix to the architecture, that is, it takes us further and further away from the original concept, which once might have been extremely creative, but now becomes only more clumsy and laden down with kluges and patches.

The backups and safeties and work-arounds we add to these systems help, but they also provide more and more opportunities for things to go wrong, often in novel and catastrophic ways. And its not just our machines; all complex systems, even human collectives like economies, governments and bureaucracies tend to require more and more a fraction of their resources be devoted to maintenance and repair functions.

The more we add on to a system or process to protect it from breakdown or inefficiency, the more likely it is that we’ll add something that will cause it to fail. This is no fault of ours, its just a consequence of the laws of complexity. The number of components increases linearly, the number of connections and relationships grows exponentially. Sooner or later, the curves have to cross. Even in the natural world, where natural selection ruthlessly weeds out things that don’t work, we occasionally have systems that fail unexpectedly. Nature’s strategy is to have them constantly evolve, but that’s an expensive response in the human ecosystem.

Older technologies often have an elegant simplicity and inherent cleverness designed into them. Modern ones tend to have a sort of Rube Goldberg quality to them, layer upon layer of redundancies and backups. This is especially the case with software.

Of course, engineers get paid to build new things, not things that last forever.