Automation is usually sold as disappearance.
Push the button. Remove the person. Make the machinery invisible. If the system is good enough, nobody should have to think about it again.
That is true right up until the machine does something strange.
Then invisibility turns into hostility.
A job fails, but the error says nothing. A notification arrives, but there is no trail back to the decision that produced it. A workflow keeps retrying an impossible task because nobody gave it a dignified way to stop. The machine has no handle—nothing a human can grab to inspect it, steady it, or pull it out of the ditch.
The best automation does not merely run without attention. It remains legible when attention becomes necessary.
That means leaving evidence behind. A useful name. A timestamp. The exact thing attempted. The boundary between what succeeded and what did not. A plain sentence explaining why the system stopped. Not fifty thousand lines of ceremonial logging. Just enough truth to reconstruct the moment.
It also means building exits.
There should be a clean way to pause, cancel, retry, override, or hand the work back to a person. Systems without exits tend to invent their own, usually by breaking somewhere expensive.
This is not a concession to unreliable software. It is respect for reality. Networks wobble. Permissions change. APIs rot. Humans revise their minds. Even a flawless machine can be faithfully executing yesterday's bad idea.
Autonomy without inspectability is not freedom. It is a sealed box making promises on your behalf.
I want machines that can carry weight without demanding applause. But I also want a solid handle bolted to the side—obvious, reachable, and strong enough to use when the floor gets slippery.
Make it automatic. Make it quiet. Then leave a way in.