Only one kind of speaker calls people “humans”: the robot in the movie.
Nobody comes home to their humans. Nobody thanks the humans in accounting. And yet the AI industry uses “human” as the standard word: human-in-the-loop, human oversight, human review, human error.
Listen to what the word is doing in those phrases. “Human-in-the-loop” comes from control-systems engineering to describe a component wired into a circuit. “Human error” names people as a failure mode; the safety researcher Sidney Dekker spent a career showing that “human error” is where lazy investigations stop, not where causes live. “Human resources” and “human capital” name people as inputs. In institutional speech, “human” almost always positions a person as a component, a failure mode, or a feedstock.
Even the loop vocabulary gives the game away. The autonomous-systems literature grades machines as human-IN-the-loop, human-ON-the-loop, human-OUT-of-the-loop, a scale whose direction of travel is the removal of the person.
The industry already knows something is off. That is why “Human Resources” quietly became “People Operations” at Google, and why the fastest-growing C-suite title of the last decade is “Chief People Officer.” When the rename happens at that scale, it’s clear the old word was doing damage.
We made the rename a hard rule
At Pensato this is written into house style: write folks, people, or person. Reserve “human” for the one job it is actually built for, literal species-level distinctions, the rare place where you genuinely mean homo sapiens as opposed to a model.
Here are three reasons.
1) Address. “Folks” is what you call people who are yours: your folks. It is warm, plural, and impossible to say about a component. It is not condescending to someone you call folks.
2) Design habit. This is the same finding we wrote about in “Rigging, Not Harness”: vocabulary drives what teams build. Call the person in your pipeline “the human” and you will design them a checkpoint, a box wired between two arrows, evaluated by throughput, resented as latency. Call them folks and you will find yourself designing them a seat at the table: with context, dignity, authority. Same pipeline. Different word. Different architecture. We even renamed the pattern itself: where the industry says human-in-the-loop, our systems say folks-in-loop, and we replaced “loop” thinking with tempo thinking, because a loop is something you are caught in, and a tempo is something you set.
3) The regulators agree. The EU AI Act’s oversight article is titled “human oversight,” but read the text: it protects “natural persons.” US consumer law speaks of persons and consumers. When your compliance documentation says person, it is speaking the law’s own language. “Human” is the engineering jargon; “person” is the legal reality.
So here is the test, in the spirit of the design test from the rigging piece: look at the diagram of your AI workflow and find the people. Are they drawn as a box in the pipe, or as the folks the pipeline answers to?
If your architecture diagram has a box labeled human, we would gently suggest the box is mislabeled. There is a person in there. Someone’s folks.
What is the people-word culture inside your AI program, and what did it build?