Nearly half of employees are shipping AI work they cannot explain, and that is a problem with a name

Rebecca Hinds has a name for a pattern she is tracking across the workforce, and the name is not gentle. She calls it “bot bullshit”: the practice of shipping AI-generated work that the person sending it could not explain if anyone asked. Her data puts that behavior at 40 to 41 percent of employees. That is not a rounding error at the margins of the workforce. It is close to half.

The figure lands hard because of what it implies about the relationship between workers and the tools they are using. The standard story about AI adoption at work is one of augmentation: people learn the tool, the tool extends their capability, and the output improves. Hinds’s number suggests something different is happening alongside that story, and possibly at much greater scale. A meaningful share of the workforce is not using AI to do better work. They are using it to produce work they cannot vouch for, and sending it out anyway.

That is a distinct phenomenon from simple error or sloppiness. Errors happen to people who understand what they are doing. What Hinds describes is a structural detachment: the human in the loop has lost the thread of what the loop produced. The work exists. The person who submitted it cannot reconstruct it. If a colleague pushes back, if a client asks a follow-up question, if the work turns out to be wrong, the person who shipped it has no ground to stand on because they did not build the ground in the first place.

40 to 41% of employees saying they ship AI work that they couldn't explain if asked. This is bot [ __ ] Rebecca Hinds

The term “bot bullshit” is pointed, and deliberately so. It extends a well-worn concept into new territory. The classic definition of bullshit, in the philosophical sense, is communication that is indifferent to truth. The speaker is not lying, exactly. They are simply not concerned with whether what they are saying is accurate. Hinds’s coinage applies that same indifference to the act of production. The employee who ships AI work they cannot explain is not necessarily trying to deceive anyone. They are indifferent to whether the work holds up, because the effort required to understand it was never made.

That indifference has organizational consequences that compound over time. Trust between colleagues depends, partly, on the assumption that the person who produced something can stand behind it. When a deliverable is a black box to the person who sent it, that assumption breaks down quietly, at first. Over many such exchanges, it can erode the baseline credibility of the work itself, not just the individual who sent it. Teams that cannot trace their own outputs cannot learn from their mistakes, cannot catch errors before they propagate, and cannot improve the process that generated them.

What makes Hinds’s framing useful is that it does not treat this as a moral failing unique to particular workers. The term names a behavior pattern, not a character type. Workers are responding to incentives. Speed is rewarded. Output volume is visible. The legibility of the reasoning behind the output is often invisible to anyone except the person who produced it, and sometimes even to them. When the incentive structure rewards shipping over understanding, a large share of the workforce will ship without understanding. That is a predictable outcome, and Hinds’s data suggests it is already the outcome.

The 40 to 41 percent figure is the kind of number that organizations may want to sit with before deciding it describes someone else’s workforce. The behavior it captures is not technically difficult to engage in. It requires only that a person prompt a model, accept the output, and send it along without the additional step of checking whether they can explain what they just sent. That additional step costs time. In environments where time is the scarcest resource, the step will often be skipped. Hinds has given that skip a name. The harder question is what organizations plan to do now that the scale of it is on the record.