This is not a case of "computers are a bit hard", it's a system that we've imposed upon ourselves that has been repeatedly demonstrated to be unsafe. because of this, we've ended up with many (many) solutions baked into software that attempt to abstract away the sharp edges of this problem from individual engineers, which leads to inconsistent assumptions about what you need to consider when writing code.

even with the best and brightest engineers there is a non-trivial probability that someone will make an assumption that is invalid based on their understanding of what can happen (ex. leap seconds never go negative!) or the library their using (this ensures monotonic time!) that could lead to disastrous results. and especially at meta scale, that probability is no longer "will someone make this mistake in our code?" but is "how many times will people make this mistake in our code?", so systemic solutions that eliminate this as a class of problem an individual can create is something we should consider.