You can also choose to avoid complexity. Often a shorter computer program that is easy to understand can do exactly what a more complicated program can. We can

You can also choose to avoid complexity. Often a shorter computer program that is easy to understand can do exactly what a more complicated program can. We can simplify interfaces between systems and ensure that their specifications are short, readable and implementable without allocation, buffers or other things that can be implemented incorrectly. We can ensure that program code is stored separately from data. Now that LLMs are getting better we could probably have them go through all our code and introduce invariants, etc. to make sure it does exactly what it's supposed to, and if it can't then a human expert can intervene for the function for which the LLM can't find the proof. I think hardware manufacturers could help too. More isolation, Harvard architectures etc. would be quite appealing.