The compounding domino model of human interaction.
A principle: every action that could have irreversible side effects outside the system must be initiated by a human.
This principle prevents automation run amok.
At the beginning, this is extremely limiting; the human must constantly be in the loop, even for mundane tasks.
But you can add layers of leverage that then actuate the layers below.
As the user gets a better handle on the quality and correctness of a given UI, they can decide to add a layer that actuates it, too.
So for example you hit "Send all" and it hits send on 10 emails in that workflow.
This gives additional leverage.
This can continue for many layers.
The key thing is 1) the human always kicks off the chain reaction, and 2) each new layer is one the human chose to put in place, with calibrated confidence in the interactions of the layer below.
This model is simple but can lead to compounding amounts of leverage.
A real-world analog is the compounding domino demonstration at science museums.
The first domino is normal sized.
But each successive domino is 10% bigger.
By the 10th domino, it's the size of a door.
The user knocks over the small domino, and it sets off a chain reaction that knocks over the door-sized one.
I originally heard this frame from Scott Jenson.