PR 3.0

Why defending your work out loud is still the definitive test of authorship, in the workplace and in the classroom

Artificial intelligence has stopped being a technological promise and become an operating layer in almost every work process. We draft with it, we synthesize with it, we design with it, we model scenarios with it. In my own case, I use it daily in both academic management and communication practice. There is nothing exceptional about that. Today, the exception is not using it.

What is exceptional — and worrying — is the uneven speed of two curves: adoption has surged, while governance moves at a crawl. And in that gap something bigger than productivity is at stake: authorship, integrity and, above all, the formation of professional judgment.

1. Fast adoption, missing rules

The data captures the gap well. Google’s Work:InProgress study found that only 30% of companies have clear policies on the use of AI, and barely 31% formally encourage their people to experiment with these tools. Microsoft, for its part, reported that 78% of those who use AI at work started on their own, with their own tools, before any corporate policy existed. And Gartner projects that by the close of 2026, more than 80% of organizations will have used generative AI models or APIs without adequate governance in place.

The phenomenon has a name — shadow AI — and a cause worth underlining: it is not born of defiance, but of friction. When the official route is slower or more cumbersome than the informal one, the policy fails before it is even written. An employee who discovers they can save two hours a day is not going to wait six months for IT to approve a tool. And it has happened to all of us at some point, hasn’t it?

In academia the picture is identical, only amplified. The 2026 Survey on AI in Higher Education in Latin America, produced by the Digital Education Council together with the Institute for the Future of Education at Tec de Monterrey and covering more than 30,000 students and faculty, found that 92% of students use at least one AI tool on a regular basis. In Mexico, the national survey by the Ministry of Public Education — with more than 1.5 million student responses — showed that 77% of university students turn to AI to produce academic texts, and that 78% of faculty use it in their own work. And the GAD3 report for Planeta Formación y Universidades, conducted in Spain, France, Italy and Colombia, found that 72% of students use AI without having received any specific training from their institution.

The institutional problem is not that students use AI. It is that they use it with no framework, no training and no one having explained to them where the tool ends and their own responsibility begins.

2. A decision that forced me to change my own parameters

A few months ago I had to make a management decision that reset my own working parameters: to regulate, through institutional policy, how far AI may be used in academic coursework. It was not a punitive decision but a formative one, and I took it from a conviction I have held for years in creative environments: research is the best raw material for sparking innovation.

AI multiplies that raw material spectacularly. It lowers the cost of exploring references, speeds up benchmarking, allows hypotheses to be tested in minutes and frees up time for what is genuinely scarce: thinking. That is its virtuous use, and regulating it badly — by banning it — would be as serious a mistake as not regulating it at all. Corporate experience proves the point: outright bans do not eliminate use; they simply drive it underground.

What does need to be regulated is something else: the traceability of authorship. That is, what may be delegated, what must be declared, what institutional information must never be entered into an external platform and — the central point — what is actually being assessed.

3. The moment of truth is not the submission: it is the defense

Here is the core of the argument. When work arrives in written form, distinguishing authorship is difficult, and automated detectors are notoriously unreliable. But when the student — or the employee — is asked to defend their proposal out loud, the scenario changes completely.

It is enough to ask why they chose that route and not another; what they discarded, and on what criteria; what would happen if one variable in the context were altered. Whoever built the proposal answers uncomfortably, but answers. Whoever merely received it, generated, improvises — and that improvisation produces a gaseous argument: fluent on the surface, hollow in its technical substance. It is not that the person knows little; it is that they never walked the path that produces such knowledge.

Nor is this a new phenomenon. More than ten years ago, when generative AI was still in its infancy, students solved their assignments by copying from the internet: it was the easiest way to shed a responsibility. And they were caught in exactly the same way, by exactly the same test: being asked to explain the meaning of a phrase or a term they clearly did not command. The tool has become infinitely more sophisticated; the test remains the same, and it remains infallible.

4. The evidence: what happens when thinking is delegated

Teaching intuition has now found neuroscientific backing. The study “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task,” conducted by Nataliya Kosmyna and her team at the MIT Media Lab (2025), followed 54 university students over four months, split into three groups — writing brain-only, with a web search engine, and with an AI assistant — measuring their brain activity through electroencephalography.

The findings are emphatic. The group that wrote exclusively with AI recorded the lowest brain connectivity, with markedly weaker activation in the networks associated with executive function, deep memory and creativity. As the sessions went on, those participants grew progressively more passive, until they were doing little more than copying and pasting the generated text. Many could not recall or quote what they had just “written.” The researchers named this effect cognitive debt: immediate convenience is borrowed at the expense of learning and long-term retention.

There are two further findings that are rarely cited and that, to my mind, matter most:

Surface without substance. The AI-produced essays scored well with automated evaluators but were penalized by human instructors precisely for their lack of depth and argumentative structure. The form was impeccable; the substance was not.

The order of use matters, and matters greatly. Those who first thought and wrote on their own, and only afterwards turned to AI to review and expand, showed a notable increase in brain connectivity — higher even than the group that worked with no tool at all.

That second conclusion is, in practice, an instruction for institutional policy: AI should not be the starting point, but the point of contrast. Judgment first, tool second.

5. The base, not the trunk

AI can draw for you, generate video, clone voices, produce complete pieces in seconds. It can hand you twenty creative routes before you finish your coffee. What it cannot do is decide which of those routes fits your brand, your audience, your reputational moment or your business objective. The essence, the meaning, the direction and the angle remain non-transferable: they are yours.

In communication this is especially critical. A piece of content can be well written and strategically useless. Formal quality stopped being a differentiator the day it became free and abundant. What sets a professional apart today is not the capacity to produce, but the capacity to decide what deserves to be produced, from what angle and for whom.

6. What every organization should regulate

From this past year of experience — and from what the data shows — I draw six minimum criteria, applicable to a company and an educational institution alike:

Explicitly permitted use. It is not enough to say “use it responsibly.” You have to define which tasks allow full assistance, which allow partial assistance and which allow none.

Declaration of use. Being transparent about AI use is not an admission of guilt; it is a practice of professional integrity, no different from citing a source.

Information protection. No sensitive, personal or strategic data should ever be entered into unapproved platforms. This is today the greatest legal and reputational risk.

Official tools. If the organization does not offer approved — and good — alternatives, every employee will choose their own. Prohibiting without providing guarantees informality.

Redesigned assessment. If the final product can be generated, assessment must shift toward the process: oral defense, justification of decisions, working logs, documented iterations.

Training, not just rules. That 72% using AI without institutional training does not need another ban: it needs someone to teach them how to ask, how to verify and how to discriminate.

7. Judgment as a personal brand asset

Let me close with a reading from communication and personal branding, the other field in which I work. We are entering a market of infinite content at zero marginal cost. In that scenario, competitive advantage shifts: it no longer lies in producing faster, but in sustaining a point of view that is your own, verifiable and defensible.

Anyone who builds a professional brand by publishing entirely generated content gains immediate visibility and deferred reputational erosion, because sooner or later the defense arrives: a meeting, an interview, an uncomfortable question from a client. And that is where the surface cracks. By contrast, anyone who uses AI to accelerate their research and then stamps their own judgment on it builds something no model can replicate: a voice with a track record, with coherence, and with the ability to defend itself.

Artificial intelligence is here to stay, and here to make our lives easier. Not to replace us. But it is worth being precise about the nuance: it does not replace those who think; it replaces — and fast — those who had stopped doing so.

In the workplace and in the classroom, the difference between using AI and depending on it has a single name. It is called judgment. And judgment, for now, cannot be generated with a prompt.

References

Digital Education Council & Institute for the Future of Education, Tecnológico de Monterrey (2026). Survey on AI in Higher Education in Latin America.

GAD3 for Planeta Formación y Universidades (2025). Report on AI use among students in Spain, France, Italy and Colombia.

Gartner (2025). Projections on generative AI governance in organizations.

Google (2026). Work:InProgress. Study on AI adoption and organizational policies.

Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I. & Maes, P. (2025). Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task. MIT Media Lab.

Microsoft (2025). Work Trend Index. Individual adoption of AI tools in the workplace.

Ministry of Public Education of Mexico (2026). National survey on AI use in higher education.