The problem is building (operating) systems that are orders of magnitude more complex than what are possible to fully understand or reason about. I don't think the top developers in the world could avoid catastrophic errors to sometimes creep into systems of that size and complexity.

Not defending Microsoft specifically, as I moved on from their operating systems to Linux 30 years ago, but I just do not see what they could hope to do. Amount of interactions to worry about will grow at least quadratic with the size of a system and there is just no way to expect human (or LLM) developers to keep up with that beyond some (very small) upper limit of system size. No matter how good the developers are and what programming languages or tools they use the result will be a house of cards of flaky components interacting in ways no one can fully predict.