Skip to content
AI strategy

Have we all become cheaters? Work in the age of AI.

Editorial collage on human work assisted by artificial intelligence.

At a glance.

  • Generative AI allows us to perform above our usual level.
  • It creates a gray area between legitimate assistance and excessive delegation.
  • Have we all, in our own way, become cheaters?

13 min read

This article was translated from its original French version with the help of AI. Read the original in French. The PDF remains in French.

This article was translated from its original French version with the help of AI. Read the original in French. The PDF remains in French.

In this publication

Until recently, cheating at work had a relatively clear meaning: lying about what we had done, taking credit for someone else’s work, pretending to know, delivering a result without actually taking the path that leads to it.

Since then, generative artificial intelligence has entered our offices.

In a few months, millions of workers have grown accustomed to asking ChatGPT, Claude, or Copilot to write, structure, code, rephrase, and analyze. The initial effect is often a rush of exhilaration. Tasks that used to take hours are done in minutes. A blank page, a difficult email, a plan that needs structure: a few prompts are enough. We feel as though we have hired a full-time assistant who never gets tired.

Using these AI tools can make us feel extraordinarily capable, in a state of “flow,” or even euphoric, as though we were getting a cognitive hit. The performance gains are real. They have been measured. They are not a fantasy. It is more than a feeling: we really do move faster.

Very soon, however, another question arises: what is work still worth when the effort that went into it becomes invisible, compressed, or outsourced to a machine? And beyond that: what happens to our moral relationship with work when we effortlessly obtain what used to take hours, sometimes years, of learning?

AI poses more than a tooling problem: it poses a problem for our “professional civilization.”

This leads us to an uncomfortable, deliberately overstated question: have we all, in our own way, become cheaters?

Working in the age of AI

The word “cheating” makes us uneasy because it seems excessive. After all, using a calculator is not cheating. Using Excel is not cheating. Using the autopilot on an Airbus A320 is not cheating. Why should asking an AI be more suspect than using any other productivity tool?

One initial answer might be that it is not necessarily cheating. It does, however, produce psychological, cognitive, and social effects that increasingly resemble those of cheating.

Why? Because cheating is not just about obtaining an illegitimate result. It is also about blurring the link between the result and the actual ability of the person presenting it. That is precisely what generative AI does on a large scale. It allows us to perform above our usual level, sometimes far above it, in a growing number of situations. It allows someone to deliver a text they could not have written alone, or not at the same speed or with the same level of skill.

That is the source of the unease. AI does more than help us work. It allows us to appear more competent, faster, more confident, more knowledgeable, and more organized than we sometimes really are. It creates a gray area between legitimate assistance and excessive delegation. And in that gray area, each of us improvises our own moral code.

Some embrace it enthusiastically. Others use it in secret. Still others promise themselves they will “just save a little time,” only to discover that they already have no desire to go back.

AI does not simply reduce work. It can also intensify it, accelerate the pace, create a kind of compulsion, and even establish psychological dependence. We keep going not just because we can move fast, but because each new prompt seems to open up another possibility.

The real issue, then, is not just laziness. It is the reshaping of our desire to work.

AI makes us want to produce because it removes some of the friction. That comes at a cost, what we might call “cognitive debt”: what we gain in immediate ease may come at the expense of what we internalize in the long run. We complete the task without absorbing the mental processes that would have allowed us to truly improve. Cognitive debt builds up.

In other words, we can succeed without learning.

Succeeding without learning

This is probably one of the most profound breaks introduced by generative AI. Throughout the history of knowledge work, effort was not just the price of producing something. It was also the very mechanism by which we improved. Writing taught us to write. Research taught us discernment. Coding taught us to think about structure. Slowness was not just an inefficiency. It was often the hidden condition that made learning possible.

AI breaks this longstanding alliance between production and training.

It allows us to obtain a result without working through the steps that once shaped the person producing it. That is why the notion of “cheating” comes up spontaneously in so many conversations. Many people have a vague sense that they sometimes produce something they have not fully earned. They also sense that the outside world mainly judges the final result. And that this disconnect between the apparent quality of the output and the actual strength of the underlying skill can become dizzying.

This vertigo brings a new form of impostor syndrome. Unlike the classic syndrome, however, it does not come from the feeling of having succeeded “by luck.” It comes from the feeling of having succeeded without struggling enough, thinking enough, or building enough ourselves. It is a new kind of imposture: not “I don’t deserve what is happening to me,” but “I no longer know exactly what still comes from me in the work I deliver.”

This is a daunting question because it touches on professional identity.

In many professions, we define ourselves not only by what we produce, but by how we produce it. Being a good lawyer, a good consultant, or a good developer is not just about handing over a sound deliverable. It rests on practiced skills, a method, judgment, a style, a way of handling uncertainty. Yet as AI takes over an increasing share of execution and even idea generation, it becomes harder to know exactly where the substance of the profession lies.

We can then understand why the debate extends beyond productivity. What is really becoming unstable is the link between work and merit.

The link between work and merit

Our organizations have long rewarded indirect signals: time spent, visible intensity, difficulty overcome, ease gained through experience.

If two people deliver a comparable result, but one worked for five hours and the other for thirty minutes with AI, what should we reward? The result alone? Skill in using the tools? The ability to verify? The ability to take responsibility?

We are experiencing a decisive phenomenon: AI disconnects visible effort from results. This is probably the source of the current confusion among managers. Organizations are no longer quite sure what they are evaluating. They see faster delivery of more polished, sometimes more uniform work. Yet they struggle to measure what those deliverables actually reflect: learning or mere delegation, mastery or mere appearances, skill building or growing dependence. This is where “workslop” comes from: work that looks sound, sometimes brilliant at first glance, but lacks substance, depth, and reliability.

The paradox then becomes cruel: someone who uses AI poorly may produce work that is slick, hollow, generic, and suspect, while someone who uses it well may be suspected of not truly deserving credit for what they deliver. In one case, AI damages quality; in the other, it damages credibility. As a result, many employees stay silent. They use the tools but conceal exactly how much they use them. They understand that performance is valued, while extensive assistance remains morally ambiguous. They therefore live with a mild form of concealment.

This is where the question “have we all become cheaters?” takes on its full meaning. Not because everyone is knowingly cheating, but because we are collectively entering a world where almost anyone can present work substantially co-produced with a machine as their own, while social, educational, managerial, and legal norms have yet to settle what that means.

In reality, we have not so much become cheaters as morally disoriented workers.

We no longer know exactly where legitimate assistance ends and improper delegation begins. We no longer know which skills we still need to possess ourselves, or which efforts still help us learn. And above all, we no longer know how to teach a profession when the machine takes over precisely the repetitive, thankless, sometimes grueling tasks that once shaped beginners’ judgment.

The shift from doing to orchestrating and then verifying looks rational on paper. The machine does more of the work; the human does more of the supervising. In the reality of organizational life, we easily forget one thing: judgment does not arise spontaneously. It develops through doing. We learn to spot an error because we have already struggled with it. We learn to sense a weakness in an argument because we have tried to build the case ourselves. We learn to supervise because we have practiced first. If young professionals begin too early to review work they could not have done themselves, we risk creating evaluators with no firsthand experience of the work they assess. Managers who have never really done the work. Experts who have never really learned.

Yet here, we could learn something from professions that have already been through a wave of automation. Faced with increasingly automated cockpits, aviation has maintained its pilots’ skills through a demanding regimen of ongoing training, simulators, and regular checks, precisely to ensure that practical skills are never lost.

This is an immense problem because it is not just about training. It is about trust.

Once no one is quite sure who can actually do what, the nature of cooperation changes. Colleagues see each other differently. Hierarchies become less stable. The outward signs of competence lose some of their value. Writing is no longer necessarily proof of thought, nor code proof of mastery. Everything is potentially assisted, and therefore potentially ambiguous.

And when everything becomes ambiguous, there is a strong temptation to take refuge in two opposing mistakes. The first is to naively celebrate augmentation: how it is done does not matter; only the result counts. The second is to condemn all AI use on moral grounds: any assisted work is supposedly diminished, inauthentic, almost impure. Neither position is sufficient.

The first disregards the fact that not all work is of equal value, and that a sound result can conceal a dangerous dependence, a loss of understanding, or a major legal risk. The second forgets that the history of work is also the history of increasingly outsourcing certain cognitive burdens, and that rejecting all assistance would be absurd. The real task is not to choose between celebration and rejection, but to redefine the boundaries of responsibility.

What remains for humans

If AI reshapes merit, it does not eliminate responsibility. Cases of AI hallucinations in legal work, emerging case law, and documented errors in critical professional settings remind us of a simple rule: the machine can produce, but it does not sign. It does not prescribe. It does not argue a case in court. It does not answer to the client, the judge, or the patient. The human user is still the one who puts their word, their name, and their profession on the line, sometimes exposing themselves to civil liability or disciplinary action.

That is why the metaphor of cheating is useful, provided we understand it properly.

Cheating here is not just “using a tool.” It means forgetting that delegation does not remove the obligation to verify, presenting work as reliable without exercising the judgment it requires, wanting the prestige of a profession without fully shouldering its cognitive and moral burden.

It would be too simple to treat this as an individual failing, because the real problem is also collective. Organizations themselves drive this confusion: they demand more, faster, more polished work, more often, without redefining their evaluation criteria. They reward output but not understanding. They deploy tools without reinventing training. They talk about augmentation without rebuilding the rules of evidence, trust, and supervision that should accompany it.

We are no longer quite sure what we should learn, demonstrate, check, or value. The old world rewarded visible effort; the new world values the ability to produce quickly amid uncertainty: neither is enough. The former can become archaic and the latter toxic. We must now distinguish several things we used to conflate: producing with help, understanding deeply, knowing how to do it again without help, knowing how to judge an output, and knowing how to take public responsibility for it.

Perhaps this is the heart of the new age of work.

The question is no longer just: “Did you do a good job?” It becomes: “What did you actually do yourself? What did you understand? What could you do again without assistance? What did you verify? What do you take responsibility for?”

This shift is demanding. All the more so in the French context. French tradition remains attached to professional craft, to know-how, to intellectual craftsmanship, to a particular idea of learning through practice. That is probably why the unease caused by AI has a particular character in France. Where some primarily see a promise of augmentation, many also perceive a threat of cognitive proletarianization: no longer knowing how to do the work, only how to configure the tools; no longer writing, only having something write for us; no longer thinking, only launching operations whose form we check superficially.

The aim is not to return to a world without powerful tools, but to avoid a world of work in which assistance and competence are persistently confused, along with speed and mastery, production and learning, fluency and truth. A society where everyone could appear able to do everything, but fewer and fewer people would still truly know how to do anything in depth.

So, have we all become cheaters?

Not really.

We have, however, entered a world where cheating has become structurally plausible, socially widespread, and morally unclear. A world where we can work honestly while feeling that we are cheating a little, where the value of work can no longer be read through its old signals, and where we will have to relearn how to distinguish assistance, delegation, competence, judgment, and responsibility.

The task, then, is neither to ban AI nor to surrender to it. It is to build a new ethic of assisted work.

  • An ethic that accepts the tool but rejects abdication.
  • An ethic that values speed, but not at the cost of understanding.
  • An ethic that accepts co-production but demands that judgment be traceable.
  • An ethic that recognizes we can produce with the machine without handing over the core of the profession to it.

Ultimately, the decisive question may not be “have we become cheaters?” but rather: will we still know how to recognize, develop, and protect what, in human work, cannot be delegated to any machine?

A society can survive with middling writing, slick slides, concise reports, and instant answers. It has a harder time surviving when it can no longer tell who thinks, understands, verifies, and who truly answers for what is produced.

This is where the age of AI becomes a test of maturity for contemporary work.

Not because it suddenly replaces everyone, but because it forces us to answer a much harder question: what do we still want to earn on our own?