Coordinate and act
Jidoka — "automation with a human touch."
Machines stop, people handle the exceptions
Build the workflow so AI runs the routine work and stops dead the moment it hits something unusual, handing that case to a person.
Borrowed from Jidoka, one of the two pillars of the Toyota Production System. Read the original source.
Changes a rule or role · An hour or more to set up · Anyone on the team can start it
Try this first
Agree one condition under which the model must stop and ask a person. Low confidence, or a value outside an expected range.
Use it when
Designing or auditing a human-AI workflow, or when the AI is quietly handling exceptions it gets wrong.
Skip it when
You need a workflow where the abnormal cases can be detected and described. If you cannot say what unusual looks like, you cannot build the stop, and the model will run straight past the cases that needed a person.
How to introduce it
Let the model run the routine, high-volume work on its own. Build it to stop and hand over the moment it meets something abnormal, uncertain, or outside its range. Run the ordinary, halt at the exceptional. Never push through.
How to show up
Map with the team where the model runs free and where it must stop. Naming the stop conditions precisely is the hard part. Push for specifics, because a vague trigger either fires constantly or never fires at all.
How long it takes
A design session to map the workflow and stop conditions. Refining the triggers continues over time.
What makes it hard
Teams set the triggers too loosely and get one of two failures: everything stops and the automation is pointless, or nothing stops and errors run through. Tightening them is the work. Watch for pressure to remove the stops once the system looks reliable, because that is when they earn their keep.
What it looks like when it's working
Routine volume flows without supervision and the cases reaching a person are the ones needing judgment. People checking everything, or bad cases sailing through, means the triggers are wrong.
How long until it sticks
Several cycles of tuning the stop conditions against real cases before the split feels right.
How you know it stuck
The team designs a new AI workflow by asking what makes it stop and hand off.
The idea behind it
This is not automate everything or automate nothing. Routine goes to the machine, judgment goes to the person, and a clear trigger moves work between them. A system that never stops has nobody watching it.
Where it comes from
Toyota Production System. Design the human-AI handoff point so the AI runs the routine and stops at the abnormal, handing judgment back to a person. Machines detect abnormalities and stop themselves so a human handles the exceptions. It separates routine machine work from human judgment. Almost eerily on-point for AI.