Picture walking through a factory where the management team is practically vibrating with excitement. Every machine is running, every conveyor is moving, and the dashboards are greener than a golf course after a fertilizer accident. Ahh...look at all that progress! Output has doubled, costs are falling, and someone has already submitted the project for an industry award.
Then you open a box coming off the line and find instructions for making more boxes.
Before you can ask who buys them, the plant manager introduces the Box Acceleration Initiative, complete with executive sponsorship, seven workstreams, and a logo. Questioning the boxes would be unhelpful because the organization has already aligned around the box roadmap. (I can practically hear someone saying, “We need to move forward.”)
Ridiculous? Absolutely, although move this operation into an office and it starts feeling painfully familiar. We manufacture presentations that generate meetings that generate actions that require presentations, and by Friday everyone is exhausted enough to feel they must have accomplished something meaningful. I find that depressing, because the effort is real even when the progress isn’t.
The business of looking busy
I think we underestimate how reassuring activity can be. A full calendar gives you something to point to, while an afternoon questioning whether the company is chasing the wrong customers can leave you with uncomfortable conclusions and nothing attractive to upload. (Try putting “realized we should stop doing this” in the weekly accomplishment tracker.)
Thinking also has a nasty habit of making things personal. Your conclusion might be that a senior leader’s favorite initiative should stop, or that six “top priorities” depend on the same three exhausted people. You can almost feel the temperature change when someone finally says it aloud, which helps explain why another workshop can seem so appealing.
Research offers an interesting clue. In a 2021 Nature paper, Gabrielle Adams and colleagues found across eight experiments that people systematically overlooked beneficial opportunities to subtract. Participants were less likely to identify useful subtraction when facing greater cognitive load or when they were not prompted to consider it. That finding makes me wince a little. Too many meetings? Form a working group on meeting effectiveness! Confusing priorities? Create a prioritization process with its own reporting requirements! Before long, the organization needs a transformation program to manage the complexity created by its last transformation program, and someone asks whether we have enough change-management capacity.
We could probably find some by stopping a few things, although I appreciate that this creates an awkward gap in the PowerPoint. (And I LOVE my beautiful PowerPoints...but sometimes the most valuable slide announces that slides 12 through 47 no longer need to exist. 🤣)
Now we can do ALL of this faster
Give our imaginary factory AI and watch what happens: instructions write themselves, boxes assemble themselves, and agents coordinate shipments while another agent prepares the executive update. The throughput is spectacular, which makes the absence of a customer almost feel like a minor administrative detail.
I use AI constantly, and I’m excited about what it makes possible. In an experiment involving 758 BCG consultants, researchers found that participants using GPT-4 completed suitable tasks 25.1% more quickly and completed 12.2% more tasks, with improved quality. On a separate task beyond the model’s capabilities, however, AI users were 19 percentage points less likely to reach the correct answer. Those numbers describe specific tasks using a particular model. The question that keeps tugging at me comes AFTER the productivity celebration. When we can produce ten strategies in the time it took to produce one, who is evaluating the assumptions? Which tradeoffs have we accepted, and which recommendations depend on customers behaving in ways we have conveniently imagined? I want the speed, absolutely, but I also want to know whether we have earned the confidence.
A 2025 study by researchers at Microsoft and Carnegie Mellon makes that question especially relevant. Surveying 319 knowledge workers, they found that greater confidence in generative AI was associated with less reported critical thinking. This was self-reported research, so it does not prove that AI causes people to lose thinking skills; it gives us reason to examine how trust affects scrutiny. And trust feels wonderful...especially when an answer arrives articulate, organized, and formatted for the leadership team. This is where I get uneasy, because the relief of having something FINISHED can make us less interested in asking whether we understand it. (Nobody wants to reopen the document five minutes before the meeting. I get it!)
Now give an agent permission to execute that recommendation, and the distance between a questionable assumption and its consequences can shrink dramatically. That possibility makes me care even more about the judgment behind the instruction.
Give thinking something useful to do
“Think harder” is lousy management advice when people are rewarded for clearing tasks and filling templates. I would start by changing a few expectations:
Make the assignment earn its existence. Before commissioning a report, dashboard, or agent, ask what decision it should improve or what outcome it should change. Get specific about the evidence that would show it worked, because “better visibility” can become an expensive subscription to information nobody uses.
Require a subtraction option. Every improvement discussion should include something we could stop, remove, or simplify.
Have an opinion before borrowing one. For consequential decisions, write down your reasoning, assumptions, and uncertainties before asking AI for a recommendation. Then let it challenge you, expose weak logic, or introduce possibilities you missed. You might completely change your mind, which is fine; understand WHY it changed. If your explanation amounts to “the analysis looked thorough,” keep going. (Yes, this takes effort. That’s the part I’m asking us to protect.)
Take the argument outside the chat window. Ask customers, inspect operational data, run a small test, or involve someone with relevant expertise. AI-generated counterarguments can help you prepare, but important assumptions still need contact with reality.
Protect some of the time you save. Revisit decisions and compare what you expected with what happened. Otherwise, every productivity gain gets swallowed by a larger workload, and we become faster at producing recommendations without becoming better at learning from them. If we never pause to compare our confidence with what actually happened, how are we supposed to get better?
Of course, thinking can become its own avoidance strategy, so please don’t turn ordering office supplies into a philosophical inquiry. Give consequential choices enough scrutiny, make small reversible decisions quickly, and keep learning. Now imagine our factory’s best employee recommends closing three production lines because customers only need what the fourth produces. The activity dashboard deteriorates, the business improves, and leadership has a revealing choice about whose performance deserves recognition. I would hate to see that person told to demonstrate more initiative.
Thinking is expensive because it can force us to disappoint people, abandon work, and admit we were wrong. AI gives us an extraordinary chance to afford more of it, provided we leave room for thinking when the machines speed up. Before approving the next factory expansion, somebody should probably open a box. 😎
References:
Adams, G. S., Converse, B. A., Hales, A. H., & Klotz, L. E. (2021). People systematically overlook subtractive changes. Nature, 592(7853), 258–261. https://doi.org/10.1038/s41586-021-03380-y
Dell’Acqua, F., McFowland, E., III, Mollick, E., Lifshitz, H., Kellogg, K. C., Rajendran, S., Krayer, L., Candelon, F., & Lakhani, K. R. (2026). Navigating the jagged technological frontier: Field experimental evidence of the effects of artificial intelligence on knowledge worker productivity and quality. Organization Science, 37(2), 403–423. https://doi.org/10.1287/orsc.2025.21838
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. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (Article 1121, pp. 1–22). Association for Computing Machinery. https://doi.org/10.1145/3706598.3713778