There is a fundamental tension between development and operations: the pace of innovation vs. the reliability of the software.
In SRE, we use error budgets to resolve this tension. It is accepted that 100% reliability is not the correct target (unless you’re building medical software or dangerous machinery), affording a small amount of unreliability in the system to allow for reasonable development velocity.
There is rarely any serious difference between software that works as promised 100% of the time versus 99.999% of the time. If we pick 99.999% as the reliability target, that affords us 0.001% as the error budget. This can be freely spent.
Often, there are other sources of failure out of our control, such as the ISPs’ networks dying which would make a higher availability on our part less valuable. E.g. if the network is 99% reliable, why bother making a 99.99% reliable service over a 99.9% reliable service?