The Machine Did 90 Percent. Why Can't the Human Walk Away?
Some products claim to have automated 90 percent of the work.
The machine generates the content, while a person checks and publishes it. The machine runs the process, then a person returns to decide whether the job is actually done. The machine has indeed done most of the work, and the person has indeed had to do much less.
Yet the person still cannot walk away.
Every time the machine stops, someone has to reenter the process, recover the context, inspect the result, and take over the final step. Automation has reduced execution time, but it has not freed attention by the same proportion.
That is what “90 percent” can easily conceal.
A percentage can tell us how many steps the machine completed or how much time it saved. It cannot tell us whether a person is now free to step away from the process. If every run still requires a handoff, someone must keep it in mind, wait for it, and remain unable to entrust the result fully to the system.
But having a person in the process is not itself the problem.
A doctor confirms a treatment plan. A responsible party approves an irreversible payment. An editor decides whether an article is ready to publish. These steps call for judgment and accountability, not repetitive labor. A person remains involved because the step belongs with a person.
Occasional exceptions are similar. When a system encounters an input outside its boundaries and knows to stop and ask for help, that does not mean the automation has failed. It means the system knows what it can do.
The question worth asking is: What does the person actually contribute upon returning?
If the person brings new judgment and assumes responsibility that a person must bear, the step is not a loose end left by automation.
If the person merely repeats the same action every time, repairs the same kind of error, or confirms a condition that could already have been specified, then “human review” may simply be a stable gap that has never been closed.
The two forms of human involvement look similar, but their value is entirely different.
The first leaves people with judgments machines should not make. The second leaves people with work the machine has not finished.
So we cannot judge the value of automation only by how much the machine does. We also have to ask what it leaves for people.
If it leaves rare exceptions, important judgments, and real accountability, human involvement is meaningful.
If it leaves the same handoff on every run, then no matter how much the machine completes, people have not truly stepped out of the process.
If the goal of automation is to free attention, the measure that matters is not the number of steps completed by the machine, but whether the normal path still requires repeated human handoffs.
When the machine has done 90 percent, do not celebrate just yet. First ask why the person still cannot walk away.