When AI Enters the Profession, What Remains of Competence?

When an artificial intelligence tool contributes to the outcome, what can still be attributed to the professional? The available studies mainly measure assisted performances, not acquired skills. It is essential to distinguish what is produced using the tool, what can be mobilized to use it, and what is learned while working with it.<\/strong><\/p>

A question from a reader<\/strong><\/h2>

After my article on professional experience as a source of skills, a reader wrote to me that artificial intelligence (AI) was shaking up the traditional paradigm of competence. The observation is stimulating. If competence is built through activity, what becomes of it when the activity is partly taken over by a tool capable of producing a text, an analysis, or a proposal in a few seconds?<\/p>

Let us recall this paradigm: competence is neither a diploma nor seniority, but the ability to interpret a situation, mobilize resources, construct an appropriate response, and assume the effects (Le Boterf, 2008; Zarifian, 1999). To say that AI shakes it up implies showing which of these elements it truly modifies. This article offers a reading, not a demonstration: the research is recent, and none directly addresses the academic recognition of experience.<\/p>

What Studies Measure<\/strong><\/h2>

Several studies have sought to measure the impact of access to an AI tool. Brynjolfsson, Li, and Raymond (2025) studied 5,179 customer support agents at a software company: the tool increases the number of problems solved per hour by 14%, and by 34% among beginners, but very little among experienced agents.<\/p>

Dell'Acqua and his colleagues (2023), with 758 consultants, observe a faster execution of about 25% for tasks that the tool masters. However, for a task outside its capabilities, consultants using it had a 19 percentage point lower chance of finding the correct solution, even though their recommendations appeared better. The tool is therefore not reliable everywhere, and it is necessary to judge what it suggests.<\/p>

These results concern the performance obtained with the tool at a given moment. They do not indicate what professionals have become capable of doing.<\/p>

An Assisted Performance is Not an Acquired Skill<\/strong><\/h2>

Three levels must be distinguished: the assisted performance, that is to say, the result obtained with the tool; the skill mobilized to formulate the problem, choose the relevant use of the tool, and evaluate the response; the developed skill, meaning the sustainable transformations in our ways of understanding and acting. Progress at the first level does not prove progress at the third.<\/p>

A study conducted in Turkey by Bastani and his colleagues (2025) with nearly 1,000 high school students in mathematics illustrates this. During exercises, access to GPT-4 improves scores by 48% with a general interface. However, when access is removed, these students perform 17% worse than those who had never had access. A version designed to guide students with hints, without giving answers, largely mitigates this effect. It would be unwise to transpose these results to adult professionals, but they illustrate a possible mechanism: a tool can enhance immediate performance without building the resources that would allow success without it.<\/p>

What the Tool Alters in Activity<\/strong><\/h2>

AI does not eliminate the gap between the prescribed and the real; it changes its form. It first shifts part of the production work to evaluation work: the professional must assess what is proposed, identify what is inappropriate or erroneous, decide whether to keep, modify, or discard it.<\/p>

It does not erase responsibility either. An assistant can suggest an interpretation or recommend a decision; the professional must then judge a proposal that contains an apparent judgment itself. A fluid and plausible response can be adopted too quickly. Reflective feedback becomes crucial: what did I understand about the situation before consulting the tool? What did I specifically ask? On what did I base my trust? The teacher communities of practice I studied allow the sharing of these criteria for judgment; the tool does not replace them.<\/p>

The Diploma and Institutions in the Face of the Tool<\/strong><\/h2>

The diploma serves as a sign: before any observation of activity, it indicates that a person is presumed to have acquired certain abilities (Boursiquot, 2014, 2023). AI can weaken its decipherability when written works, produced without observation, can be partly assisted. It also revives a question I raised in a previous article: does society recognize the diploma or the competence?<\/p>

Institutions are beginning to respond. Canadian federal research funding agencies require the declaration of uses of generative AI, verification of information, and assumption of responsibility. Laval University regulates the declaration of use in memos and theses. In Haiti, Philippe (2026), in these columns, pointed out the difficulty of identifying explicit regulation of the academic use of AI. When rules are lacking, evaluation depends on the implicit expectations of teachers and committees. The Orélien affair in Quebec finally shows the fragility of evidence: the use of a tool, or its absence, can only be demonstrated by stylistic clues.<\/p>

These choices rely on representations. If competence is viewed as the ability to produce a result alone, the tool makes it less distinctive. If it is viewed as the ability to interpret, judge, and assume a decision, the tool makes it more visible. The paradigm is not overturned; it is put under tension.<\/p>

What This Changes for the Recognition of Experience<\/strong><\/h2>

During a validation of acquired experience (VAE), the jury never has direct access to lived experience, only to the traces and verbalization proposed by the candidate (Vermersch, 1994). However, a structured and well-written text no longer guarantees that the candidate is the author of the analysis they present, nor that they understand it.<\/p>

Two symmetrical risks appear: accepting a file as proof whose quality mainly depends on the tool, or a priori suspecting any well-written file and penalizing those who legitimately use it to express real experience. The jury can question the candidate about the described situations, the choices made, the alternatives discarded. Support becomes more necessary: it is about helping the candidate recognize what they can genuinely do, not producing the evidence in their place.<\/p>

Finally, mastering a tool is sometimes a valuable skill, but it is not necessarily the one certified by a diploma. The certifier should specify the expected autonomy: some situations require acting without assistance, while others require reasoned mastery of instruments that have become ordinary in the profession.<\/p>

What Needs to Be Reconsidered<\/strong><\/h2>

AI does not render competence obsolete. It compels us to re-examine the conditions of its construction, exercise, and recognition. An assisted performance can mask a lack of learning; an activity organized around verification and reflective feedback can also open opportunities to learn. It would be premature to take for granted that human judgment will be strengthened: it remains an object of inquiry itself.<\/p>

For certification institutions, the response cannot rest on the appearance of documents. It must focus on the situations, choices, justifications, and real requirements of the diploma.<\/p>

This text is an abbreviated version of an article published on boursiquotherve.com, which contains the complete references.<\/i><\/p>

Hervé Boursiquot holds a PhD in education sciences and is the author of Diplomas, Skills, and Validation of Acquired Experience in Haiti<\/u> (L'Harmattan, 2023) and Teacher Communities of Practice: Challenge of Raising the Quality of Education in Haiti<\/u> (L'Harmattan, 2024).<\/i><\/p>