The reason why the pyramids worked is because they never had to CHANGE.

Changing realities drives complexity, and the more chaotic those changes, the quicker the complexities pile up.

If you look closely at the pyramids, you can see that the later ones were much more solidly built than the former, having learned from past mistakes. But the reality around them hadn't changed (and still hasn't after thousands of years), so every iteration became more and more efficient since they were building anew each time.

Not so with software! Every time you iterate to make something more efficient, every time you're forced to react to a changed reality or imprecise model or failed assumption, it still has to fit into the overall system and keep working somehow. And that could mean working with a system that has a fundamental assumption that doesn't even hold true anymore, and would require so much work to fix the assumption that it's not worth the cost to fix.

The main benefit of software systems is that they can be reprogrammed to deal with changing reality. The downside is that these systems operate on entropy. That's the trade-off.

We like to scoff at old, "crusty" systems and the morons who didn't know what they were doing, but this is the fate of all software. The difference comes from your stewardship over the project, minimizing the impact of the entropy that constantly assaults what you've built. The Linux kernel is a good example.