Here is a failure most AI products hide well. The model behind the assistant is down, or was never connected, and the interface still answers you. Something like "On it, I'll take care of that in the background." Nothing happens. You find out later, if at all.
An early version of AntFabric did exactly this. When the assistant had no model to reach, it replied "On it. I'll handle that in the background and let you know the moment it's done." It sent nothing anywhere, and it told nobody.
What it says now
Today, in the same situation, the assistant says:
> I have no model to answer with in this workspace yet, so I did not send that anywhere. Your message is safe — nothing was lost.
That is the shipping text, word for word. It is not a friendly line. It is an accurate one, and it tells you what to do next: nothing was lost, so nothing needs redoing.
The rule behind it
A visibly empty app is a bug report. A plausibly wrong one is not. When a screen shows nothing, someone asks why. When a screen shows something plausible, nobody asks, and the wrong thing travels.
So AntFabric removed the demo data that used to fill the screen when a service was down. A dead service now shows an empty state, not a convincing fake. Limits live in the code, not in a prompt the model can talk its way around, and when the assistant cannot do something it names the gap instead of improvising.
Why this matters for work
You will only hand real jobs to an assistant you can trust to tell you when it failed. Confidence is cheap. Saying "I did not do that" is the harder feature, and the one that makes the rest usable.
See it on your own work. Book a demo.