Reliability as Availability

A system is available if it works when you need it. An AI service is available if it is not down for maintenance. A colleague is available if they show up to work and answer their email. Availability is about presence and responsiveness.

Availability is useful. It is not trust.

But here is where availability becomes interesting in an organizational context. When you delegate something to another person, you are implicitly trusting them to do more than show up. You are trusting them to care about the outcome. You are trusting them to speak up if they see a problem. You are trusting them to prioritize the work in a way that aligns with what actually matters, even if you did not specify exactly what that is.

An AI system can be available in the sense that it is running and ready to respond. But it cannot care about whether the task actually matters. It cannot prioritize among competing goods. It cannot raise its hand and say, "Wait, I think we should reconsider whether we should be doing this at all." It will not stay late because the deadline is tomorrow. It will not cut a corner differently than specified because that corner-cutting would actually cause harm.

This does not make AI unsuitable for execution. It makes AI unsuitable for anything that requires judgment about what matters.