As generative Artificial Intelligence (AI) makes polished work easier to produce, leaders must learn to distinguish genuine capability from its convincing appearance.
Imagine walking into a room wearing a Rolex Daytona. From across the table, everything looks right: the proportions, the ceramic bezel, the weight of the bracelet. Someone notices it. Someone compliments your taste and quietly revises upward what they assume about your career or success.
But one person in the room knows with certainty that it is not a Rolex: you.
That private discomfort resembles the anxiety surrounding irresponsible use of generative AI. The presentation is polished. The report sounds intelligent. The strategy deck contains the right frameworks. Your colleagues are impressed. Yet somewhere inside, you know you could not have produced it yourself, and you may not fully understand what you have presented.
As AI improves, that gap between appearance and substance will become harder to see and more important to confront.
WHEN OUTPUT STOPS PROVING COMPETENCE
Modern counterfeit watches can be convincing because they reproduce many of the signals visible to an outside observer. The philosophical problem is not simply that an imitation exists. It is that the distance between the genuine article and its appearance has become small enough to create information asymmetry.
AI is creating a similar problem in knowledge work. The spreadsheet works. The essay reads beautifully. The code runs. The proposal uses the language of an experienced consultant. Nobody necessarily knows how much of the intellectual machinery came from an AI system. Perhaps nobody cares until the questions begin.
Why did you choose this model? What assumption drives the forecast? What evidence contradicts the thesis? What happens when the client’s constraint changes? Eventually, someone opens the caseback.
We have long treated a finished artifact as evidence of the process behind it. A well-reasoned article implied that someone had read, discarded weak arguments, and struggled toward a conclusion. A financial model suggested that its author understood the assumptions underneath it. Working software implied some knowledge of programming.
Generative AI weakens that inference. The output may still be excellent, but output alone no longer proves competence. We can now acquire many outward symbols of intellectual accomplishment without building the internal models that once produced them. The applause can remain real even when the capability is borrowed.
AI CAN BUILD REAL CAPABILITY
An easy anti-AI argument fails here because AI can create genuine capability. In a randomized experiment published in Science, Shakked Noy and Whitney Zhang found that professionals using ChatGPT completed certain writing tasks 40% faster while their output quality rose by 18%. A Harvard Business School field experiment with Boston Consulting Group found that consultants using GPT-4 finished appropriate tasks about 25% faster and produced markedly better work.
The qualification matters: appropriate tasks. The Harvard researchers described a jagged technological frontier. AI performed extraordinarily well in some areas and proved unreliable just beyond them. Competence therefore includes knowing when the machine has crossed that frontier.
Watchmaking offers a useful comparison. A modern Rolex is not produced through romantic, preindustrial craftsmanship. Skilled watchmakers work alongside automation, advanced metallurgy, proprietary machinery, laboratories, and rigorous testing. Technology did not eliminate the craft; it became part of the craft.
The relevant question is not whether a machine helped. It is what the machine allowed the person to become capable of doing.
Two people might ask AI why the Daytona matters. One copies 10 reasons into a social media post. The other keeps asking: Why does a column-wheel chronograph matter? What are the trade-offs between finishing, robustness, and serviceability? Why can provenance make one reference more valuable than a technically superior watch? Then that person reads independent sources, challenges the answers, and develops an informed view.
Both used AI. Only one outsourced curiosity. The other accelerated it. AI becomes corrosive when it ends inquiry and powerful when it generates the next question.
PROTECT THE FRICTION THAT TEACHES
A 2025 MIT Media Lab preprint offers an early warning. In a small study of essay writing, participants using a large language model showed weaker brain connectivity than participants who wrote without tools. They also recalled less of their own work and reported less ownership of it. The paper has attracted methodological criticism and should not be treated as a final verdict, but it asks the right question: What happens when producing an answer repeatedly replaces constructing one?
The answer is not to preserve inefficient work for its own sake. Accountants need not abandon spreadsheets. Architects need not reject computer-aided design. Writers need not surrender spellcheck. AI should remove transcription, formatting, routine correspondence, first-pass summaries, and other cognitive drudgery.
The harder task is distinguishing wasteful friction from formative friction. Reading 30 papers to find three useful insights sounds inefficient, yet the 17th paper may give a researcher an intuitive command of the literature that no summary can supply. Writing a difficult first draft may reveal that an argument is not yet understood. Debugging software by hand can build the mental models that later help an engineer recognize bad generated code.
Some inefficiencies are apprenticeships in disguise. We should automate the friction that consumes human capacity while protecting the friction that creates human capability.
OPEN THE CASEBACK
AI also changes where intellectual advantage lives. When almost anyone can generate a plausible answer, the scarce skill becomes judgment: knowing what to ask, which assumption is missing, what would falsify a claim, and when the machine is confidently wrong.
For leaders, the ethical boundary is not whether someone used assistance. Companies employ researchers. Writers work with editors. Executives rely on teams. The boundary is whether we create a false impression of capability where that capability materially matters.
Using AI to format a proposal compromises little. Selling a cybersecurity architecture to a bank without understanding its threat model compromises a great deal. Improving the grammar of an academic paper is different from outsourcing the intellectual work that the degree is meant to certify.
A practical discipline follows. Use AI to remove drudgery, but retain responsibility. Interrogate any claim important enough to repeat. Never outsource the final judgment. Occasionally work without the tool, not as punishment but as calibration. Can you still explain the concept, defend the decision, and recognize when the answer is wrong?
Rolex famously linked its Oyster watch with swimmer Mercedes Gleitze’s 1927 English Channel feat. The story worked because the object pointed back to an achievement. The watch was a signal, but there was something underneath the signal: engineering, testing, and human endurance.
Our use of AI should meet the same test. Let the machine make us faster, expose us to unfamiliar fields, challenge a thesis, and clear away boilerplate. Then open the caseback. Do I understand this? Could I defend it? Did the tool sharpen my judgment or merely improve its appearance?
In an age when almost anyone can produce an answer, we may learn to value something deeper: how carefully a person thinks, how intelligently they question, and whether, after the machines have done their work, there is still a capable human being underneath.
Even if nobody else opens the caseback, you still know what you are wearing.
Jake Ervin Jonathan C. Go is a member of the MAP NextGen Committee, the managing partner and CEO of Springboard Philippines, a professor at De La Salle University, and author of Careless Machines: 24 Rules to Keep Your Judgment in the Age of AI.