Your AI Is Producing Work. Your Employees Are Babysitting It.

The productivity promise of AI was straightforward: machines do the grunt work, humans do the thinking. What's emerging instead is a third category of labor — humans watching machines do the grunt work. That category now has a name, "botsitting," and a price tag: more than 6 hours of employee time per week, consumed not by producing outcomes but by verifying that the AI produced the right ones.

Six hours a week is not a rounding error. It's 15% of a standard 40-hour week, every week, per employee. For a 500-person knowledge-work organization, that's the equivalent of roughly 75 full-time employees doing nothing but checking AI output. Before you reach for a re-training budget or a new prompt engineering workshop, ask yourself: is your AI deployment actually reducing labor costs, or has it simply shuffled them into a column that doesn't show up on a headcount report?

The verification instinct is not irrational. International SOS research cited by Personnel Today found that almost 60% of senior decision-makers believe AI's greatest value lies in verifying or validating information from multiple sources — which sounds sensible until you recognize the implication: if the primary value of AI is checking other information, and humans must then check the AI, you have built an expensive loop, not a productivity engine. The checking never terminates; it just migrates upward.

This is where the ROI timeline problem bites hardest. Economists studying AI adoption have consistently warned that measurable returns lag deployment by years — HR Executive's coverage of economist commentary notes that AI ROI could take years to materialize in HR functions specifically, even as investment pressure mounts quarter by quarter. Most organizations are currently in the worst possible position: they've absorbed the change management costs, the licensing fees, and the workflow disruption — but they haven't yet reached the inflection point where institutional learning about how to use the tools actually pays off. The botsitting hours are a tax levied precisely during that gap.

What makes the botsitting phenomenon strategically significant — rather than just operationally annoying — is what it reveals about how AI has been deployed. Most enterprise AI rollouts have been grafted onto existing workflows rather than used to redesign them. The tool gets inserted into step four of a twelve-step process; steps one through three and five through twelve remain unchanged. Employees rationally treat the AI output as a draft from a junior colleague they don't fully trust yet, which means verification becomes a professional norm, not an exception. Once that norm calcifies, it's nearly impossible to walk back without explicit organizational intervention.

MIT Sloan Management Review's coverage of the AI Atrophy problem surfaces the inverse risk: organizations that stop checking AI output risk eroding the critical thinking skills required to know when something has gone wrong. CIOs at the 2026 MIT Sloan CIO Symposium flagged the erosion of judgment as a direct consequence of over-delegation to AI systems. So the dilemma is genuine — check too much, and you pay the botsitting tax; check too little, and you hollow out the cognitive capacity that makes checking meaningful in the first place.

McKinsey's analysis of where AI value actually flows puts a finer point on the escape route: the next wave of returns goes to organizations that redesign business models and eliminate friction rather than automate existing steps. That framing reframes botsitting not as a training problem or a trust problem, but as an architecture problem. The question isn't whether employees should verify AI outputs — it's whether the workflow was designed with verification costs priced in, and whether those costs are lower than the alternative.

Senior HR and operations leaders who are currently measuring AI success by adoption rates and feature utilization are looking at the wrong instrument panel. The number that matters is unbillable verification hours per employee per week. If that figure is climbing — and the evidence suggests it is — then your AI program is not yet a productivity investment. It is an expensive redistribution of labor that has made the work less visible without making it less present.

The question worth taking into your next board conversation: if your employees are spending 6-plus hours a week auditing AI, what exactly did you buy?