When the Output No Longer Proves Competence: AI Is Changing the Value of Human Skill

For decades, we have used the output of work as one of the main ways to assess a person's competence. A developer produced code, an analyst prepared an analysis, a manager wrote a report, an engineer performed calculations, and a candidate presented work demonstrating what they could do. The output was rarely perfect proof of competence, but it generally remained a reasonable indicator of the skill required to produce it. Artificial intelligence is beginning to weaken that relationship.

A person can now produce an excellent report without fully mastering the subject, generate code they would be unable to explain in detail, carry out a sophisticated analysis without knowing some of the methods used, or prepare a convincing presentation based on reasoning largely built by an artificial intelligence. The output may be perfectly valid and useful to the organization. It nonetheless tells us less and less about the skills actually held by the person presenting it.

Academia is already encountering this difficulty. MIT observes that artificial intelligence can produce credible answers to virtually every traditional form of written assignment, from essays to mathematical problems, proofs, and programming work. This shift is leading the institution to consider other ways of assessing students' real mastery, notably through demonstrations, interactions, skills evaluations, and situations in which the student must explain and defend their understanding.

The world of work should watch this transformation very closely, because the same phenomenon is beginning to take hold there. When the quality of a deliverable no longer makes it possible to infer directly the competence of the person who produced it, companies must rethink how they recruit, train, assess, and develop their teams.

An even more important question then arises: as AI performs a growing share of the work, which human skills must we absolutely continue to develop and preserve?

Artificial intelligence is already increasing the productive capacity of many knowledge workers. An employee can write faster, analyze more information, program with assistance, prepare scenarios, summarize documents, and obtain an almost instantaneous first answer to problems that previously required several hours of research.

This increase in capacity is a tremendous opportunity for organizations. It allows people to devote more time to complex problems, decisions, creativity, human relationships, and activities where their experience produces more value. The issue is certainly not to artificially preserve every task that machines can perform efficiently. It is, however, necessary to distinguish the task we can delegate from the competence we must retain.

Take a developer who uses artificial intelligence daily to produce code. That use can considerably increase their productivity. If they retain sufficient understanding of the architecture, the algorithms, the dependencies, the security, and the behavior of the generated code, AI augments their professional capability. It allows them to work faster while using their expertise to steer, verify, and integrate what is produced.

The situation is different if, over time, the developer progressively loses the ability to sufficiently understand the code they accept. The same productivity gain can then mask a decline in the human capability available within the organization.

The phenomenon can appear in virtually every intellectual function. An analyst can produce more analyses while practicing certain fundamental methods less. A cybersecurity professional can quickly obtain recommendations without developing the same diagnostic depth. A manager can receive extremely well-structured summaries without going through the information that would once have contributed to their understanding of the file. A new employee can quickly become productive thanks to a copilot while more slowly acquiring the knowledge that will eventually allow them to exercise judgment independently. We might call this phenomenon a skills debt.

Like technical debt, it can be perfectly rational when it is understood and managed. An organization has no reason to manually maintain a skill for every automatable task simply because it was once necessary. Technologies have always transformed professions and made certain knowledge less important while other knowledge gained value. The debt arises when the organization progressively stops knowing which skills it is losing.

A company may then observe an increase in its productivity and conclude that its operational capability is increasing in the same proportion. Yet the two are not necessarily equivalent. It can produce more while simultaneously becoming more dependent on certain models, vendors, agents, or automated systems.

This distinction deserves a place in discussions about AI adoption. Traditional indicators easily measure the number of cases handled, time saved, volume of code produced, or productivity gains. It is far harder to measure the understanding that was not developed, the experience that was not acquired, or the skill that gradually eroded.

The issue becomes particularly important for new employees. An experienced person adopting a copilot often has several years of knowledge allowing them to recognize an odd answer, understand the trade-offs proposed, and detect a subtle error. A beginner learning their profession directly with the same tool can achieve comparable results much faster without necessarily building the same foundations.

This situation creates an organizational paradox. Artificial intelligence can considerably accelerate the rise in productivity while complicating the rise in competence.

Part of professional training will therefore have to be rethought. The objective should no longer be solely to teach employees to perform the tasks that currently make up their job. It will also be necessary to determine which fundamental knowledge will allow them to understand, supervise, and challenge the work performed by the systems they use.

MIT is encountering a very similar version of this question in teaching. Some courses deliberately continue to require students to build their own foundations before relying more heavily on artificial intelligence. In programming, for example, the objective remains to build enough independence to allow future professionals to properly evaluate the work of other people and of AI tools.

This thinking can be directly transposed to the enterprise. We do not necessarily need every professional to continue manually performing all the operations that AI can take on. We do, however, need to determine the level of understanding below which delegation becomes a dependency that is difficult to supervise. This also transforms the assessment of skills.

If a candidate presents an excellent document prepared with AI assistance, that document may demonstrate that they know how to use the tool effectively, which is now a relevant skill. It does not necessarily demonstrate that they master all the knowledge contained in the document. Companies will therefore have to learn to distinguish the ability to produce with AI, domain understanding, and the ability to exercise judgment when the machine gets it wrong.

Interviews can progressively incorporate more interactive problem solving, explanation of reasoning, defense of a recommendation, or critique of a generated answer. Professional assessments can ask a person to explain why a solution works, identify what could fail, or choose among several proposals produced by an AI. The final output remains important. The process for evaluating it simply becomes richer.

This evolution can even improve certain talent management practices. We have long confused the ability to produce a deliverable with mastery of the problem that deliverable addresses. Artificial intelligence forces us to look deeper and to ask which skills we are really seeking to develop. This question becomes even more strategic when linked to business continuity.

Organizations already have plans for operating when a server fails, when a data center becomes unavailable, when a critical supplier experiences an outage, or when a cybersecurity incident requires isolating part of the infrastructure. They plan for backups, recovery mechanisms, redundancies, and degraded-mode operating procedures. The massive integration of artificial intelligence progressively introduces a new category of dependency: cognitive dependency.

Imagine an organization in which a significant part of programming, analysis, technical support, documentation, internal research, and decision preparation rests on AI assistants and agents. That organization can become remarkably high-performing. But what happens if some of those systems become temporarily unavailable?

An outage is only one scenario among several. A vendor may change its service, a vulnerability may force the company to temporarily suspend a model, a regulation may restrict a use case, a confidentiality issue may require shutting down an integration, or a cybersecurity incident may demand the immediate isolation of several agents. The continuity question then becomes very concrete: do the humans who remain still know enough about how the work functions to maintain essential operations?

We might speak here of cognitive continuity: an organization's ability to preserve the knowledge, skills, and judgment needed to continue its critical functions when certain artificial intelligence capabilities become unavailable, unreliable, or inappropriate.

This continuity does not mean maintaining two parallel organizations, one automated and the other entirely manual. That would be costly and often pointless. It calls instead for identifying the genuinely critical skills and determining the minimum level of human capability that must be preserved.

In some activities, an AI interruption will simply cause a temporary drop in productivity. In others, it could prevent the organization from making a decision, diagnosing an incident, or understanding a critical system well enough to intervene. These situations do not carry the same level of risk and should not receive the same treatment.

Skills debt then becomes measurable through the risk it creates. A skill whose disappearance has virtually no consequence can naturally be abandoned. A skill necessary for supervising a critical system, for security, for continuity, or for the ability to regain control deserves, on the contrary, to be maintained.

This approach profoundly changes the conversation about the future of work. It is no longer simply a matter of determining which tasks will be performed by humans and which will be entrusted to machines. It is also necessary to determine which human capabilities must survive the automation of those tasks.

For executives, IT leaders, HR, and transformation teams, this implies far closer collaboration. Process mapping should progressively be accompanied by a mapping of skills and cognitive dependencies. When a capability is transferred to AI, the organization should know which knowledge becomes less used, which remains necessary to supervise the machine, and which gains value precisely because automation is increasing.

This thinking also connects with cybersecurity and resilience. An organization cannot be considered truly resilient merely because its systems have backups and technical redundancies. It must also retain enough human knowledge to understand what is happening when those systems behave differently from what was expected.

From a Hypersecurity perspective, resilience thus concerns the infrastructures, the data, the identities, the technologies, and the human capabilities that allow the organization to retain its decision-making autonomy. A technological dependency can be perfectly acceptable when it is known, chosen, and accompanied by a plan for managing its consequences.

Artificial intelligence therefore invites us to broaden our definition of competence. In many professions, being competent will no longer necessarily mean knowing how to carry out every step of the work alone. Competence may include the ability to properly frame a problem, orchestrate tools, understand their results, identify their limits, verify the critical elements, intervene when they fail, and own the final decision.

This evolution does not necessarily impoverish human skill. On the contrary, it can raise it a level, provided we consciously choose the capabilities we want to develop and those we can entrust to machines.

Artificial intelligence makes a considerable increase in human and organizational capacity possible. People can produce faster, explore more possibilities, and access almost instantly capabilities that previously required far more time or expertise.

This transformation does, however, make the output less revealing of the competence behind it. An excellent report, a sophisticated analysis, or a working program can now result from a variable combination of human knowledge and artificial capabilities. Evaluating only the final product is progressively becoming insufficient to understand what a person or an organization can really do.

Companies will therefore have to learn to manage two forms of capital simultaneously: the technological capabilities they acquire and the human skills they choose to preserve. Some knowledge will naturally become less important. Other knowledge will have to evolve toward supervision, judgment, integration, and decision-making. Some will have to be maintained precisely because it constitutes a fallback capability when automated systems are no longer available or trustworthy.

Skills debt and cognitive continuity could thus become important dimensions of transformation through artificial intelligence. They make it possible to move beyond productivity measurement alone and ask a far more strategic question: after several years of AI adoption, what do we want our organization to still be capable of understanding and accomplishing on its own?

For Quantum Beyond, this thinking fits directly into an approach where the adoption of artificial intelligence aims to strengthen teams and develop organizational capabilities. AI Adoption, AI Organizational Transformation, AI Workforce, and AI Governance must allow companies to take full advantage of machines while understanding how knowledge, responsibilities, dependencies, and essential human skills are evolving.

The next stage of AI adoption will therefore not consist solely of learning to delegate more. It will consist of learning to delegate consciously: knowing which tasks can be entrusted to machines, which skills must evolve, and which human capabilities must remain strong enough to understand, supervise, and regain control when the situation demands it.

An organization genuinely augmented by artificial intelligence should not only know what its machines are capable of doing. It should also know what its humans must still be capable of doing without them.