Your Team Stopped Checking the AI’s Work. You Taught Them To.
Researchers recently ran a simple experiment that should unsettle anyone rolling out AI. Emma Wiles, a professor at Boston University’s Questrom School of Business, working with colleagues from Boston Consulting Group, recruited 1,261 managers, directors, and HR and finance executives across the United States, Canada, and the European Union. Each was given the same set of documents to review, each seeded with the same errors. The only thing that changed was the label. Some reviewers were told the work came from a chatbot. Others were told it came from an “AI employee” with a name and a defined role.
The group reviewing the “AI employee” caught 18 percent fewer errors. The same mistakes, the same documents. All that differed was what the reviewer had been told about who or what produced the work.
Sit with that number, because it is not really a story about AI. It is a story about accountability. When the work carried a human-sounding label, managers quietly concluded that catching its mistakes was someone else’s job. As Wiles put it, they could always blame the tech team or the executives who wanted an “AI employee” in the first place. The tool did not get worse. The humans stopped owning the output.
It has been pointed out to me that a caveat may apply to AI-generated code. Coding is very different from, say, marketing or standard research. In coding, there is a “correct” solution (bug-free, all requirements met). Therefore, having a human review the work may be less effective than system-driven changes at improving AI accuracy.
Leaders Are More Complicit Than They Think
That last point is where leadership comes in, and where most of us have been looking the wrong way. The early corporate story about AI has been about efficiency and speed. Cut heads. Compress timelines. Mandate the tools and measure adoption. That story quietly signals to employees that the goal is to produce more, faster, with AI doing the heavy lifting. It rarely says a word about who remains responsible for the result.
Employees hear that signal clearly. When leaders push AI everywhere without standards for what quality of work is acceptable to submit, they get exactly what they asked for: more output, less ownership. Researchers at BetterUp Labs and Stanford’s Social Media Lab coined the term “workslop” for the resulting product, defining it as AI-generated work that looks polished but lacks the substance to move a task forward. In their survey of more than 1,100 employees, 40 percent had received workslop in the past month. Each instance took the recipient nearly two hours to untangle. Recipients also rated the sender as less capable and less trustworthy.
Notice the pattern. In the Boston University study, managers stopped catching errors because the framing let them off the hook. In the workslop research, employees pushed unfinished work downstream because no one had told them the standard still applied. Both are failures of accountability, and accountability starts at the top.
The Quiet Surrender of Judgment
Beneath the accountability problem lies a deeper one. People are handing their judgment to the machine. A 2025 study by Microsoft Research and Carnegie Mellon surveyed 319 knowledge workers and found something counterintuitive: the more a worker trusted the AI, the less critical thinking they did. The more they trusted their own ability, the more they thought. Confidence in the tool bred passivity. Confidence in oneself bred scrutiny.
This is what psychologists call cognitive offloading, and it is not new. Socrates worried that writing would erode memory. What is new is the scale and the seduction. People have stopped drafting, editing, and wrestling with a blank page. They let the tool create a false sense of rigor in their research, self-evaluations, and reviews of colleagues, and they stop reading closely enough to catch the errors the machine confidently produces.
Contrast that with how serious leaders have always treated the work of the mind. General James Mattis warned that the officer too busy to read will learn the hard way, at a cost measured in his own troops. The point was never that reading is virtuous. It was that judgment is built, not downloaded. You cannot outsource the very judgment that is supposed to make the output worth trusting.
What Leaders Should Actually Do
Start by reversing the sequence. Ask people to think first, then reach for the tool, not the other way around. Set explicit boundaries for where AI belongs and, just as important, for what quality of work is acceptable to submit as finished. Hold AI-assisted output to the same standard as human work, and make it unmistakably clear that a person, not a named “AI employee,” still owns every result.
Then rethink the jobs themselves around a human-in-the-loop model, and ask a harder question in the process: which skills actually matter in this role now, and how do we build the judgment and context that make AI useful rather than dangerous? Workers who think are the ones who are confident in their competence. The path to better AI runs directly through more capable people, not fewer of them.
Finally, understand the tool yourself. You cannot credibly demand that your organization find efficiencies with a technology whose limitations you have never bothered to learn. Some leaders boast about eliminating entire functions with AI, only to discover later that efficient and effective are not the same. Festina lente, the Romans said. Make haste slowly. That old instinct has rarely mattered more.
The Honest Part
I will admit the pull is real. I have gone down the AI rabbit hole myself this year, and I have caught myself accepting a clean, confident paragraph without asking whether it was right or truly mine. The convenience is seductive precisely because the output looks finished. That is the trap, and no one is immune, least of all the leaders who set the tone.
So here is the question I would put to you. When you rolled out AI, did you make it clear who still owns the work? Your people have already answered that question, whether or not you meant to ask it.
Angelo Santinelli is the founder of Entrepreneurial Edge Executive Coaching and Advising and a strategic advisor to PE/VC-backed and founder-led companies. He works with CEOs and executive teams on strategic execution, leadership development, and organizational performance.
References
1. Wiles, E., Hsu, M., Bedard, J., & Kropp, M. (2026). Putting AI on the Org Chart: Evidence on Delegation and Oversight [Working paper]. Boston University Questrom School of Business and the MIT Initiative on the Digital Economy. Summarized in Kropp, M., Bedard, J., Wiles, E., Hsu, M., & Krayer, L. (2026, May 6). Research: Why You Shouldn’t Treat AI Agents Like Employees. Harvard Business Review. https://hbr.org/2026/05/research-why-you-shouldnt-treat-ai-agents-like-employees
2. Niederhoffer, K., Kellerman, G. R., Lee, A., Liebscher, A., Rapuano, K., & Hancock, J. T. (2025, September 22). AI-Generated “Workslop” Is Destroying Productivity. Harvard Business Review. https://hbr.org/2025/09/ai-generated-workslop-is-destroying-productivity
3. Lee, H.-P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers. Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3706598.3713778
4. Mattis, J. N., & West, B. (2019). Call Sign Chaos: Learning to Lead. Random House.
5. Plato. (c. 370 BCE). Phaedrus.
6. Suetonius. (c. 121 CE). Divus Augustus. In The Twelve Caesars.
