Fallback

The plan B, agreed in advance. If the AI you normally use is down or overloaded, the work goes automatically to a second one instead of stopping. Same idea for answers: if the AI isn't confident, it falls back to a safe standard reply, or to a person.

Your backup supplier. You already keep one for when the usual merchant is out of stock — you don't wait for the shortage to start ringing round. A fallback is exactly that, wired in ahead of time.

Why it matters: AI companies go down, and they don't warn you first. If your setup can only talk to one of them, their bad afternoon is your bad afternoon and there's nothing to do but wait. Which is why we always build with a fallback and never lean on a single company. It's a fair thing to ask before signing anything: what happens to this when the AI company behind it goes down?

Related: Model · Guardrail