Every age believes it understands the meaning of its most important words, even as the realities they describe quietly change beneath them. “Freedom”, “responsibility”, “authority” and “trust” have survived centuries of transformation, yet none means precisely what it meant to previous generations.
“Competence” may have undergone a similar transformation, although one particularly difficult to recognise because we continue to use the word with such confidence. We describe physicians, judges, teachers, engineers and leaders as “competent” as though the quality were self-evident and essentially unchanged. Yet the characteristics through which competence is recognised have shifted considerably. Now, AI may be exposing just how far that shift has gone.
For much of professional history, competence revealed itself most clearly when certainty became impossible. The competent physician possessed medical knowledge, but the difficult patient was precisely the one for whom evidence and established practice did not produce an obvious answer. The competent judge knew the law, but judgment became necessary where applying a general principle to a particular human situation required interpretation. The competent leader understood the organisation, but leadership mattered most when precedent could no longer determine what should happen next. Competence therefore contained a common element: the ability to act responsibly when rules, knowledge and experience had carried the professional as far as they could.
Knowledge remained the foundation upon which judgment rested. Yet professional competence included an acceptance that reality would occasionally exceed whatever systems had been designed to describe it. Expertise was valuable not because the expert always possessed an answer, but because the expert could recognise when an answer remained uncertain and still assume responsibility for deciding what should happen next.
Modern institutions had good reasons to become uncomfortable with such an elusive definition. As organisations grew larger, professions more specialised and decisions more consequential, institutions needed to identify competence, certify it, reproduce it and evaluate it at scale. Medical education developed standardised examinations and professional guidelines; corporations created performance metrics and competency frameworks; universities formalised assessment; regulators demanded demonstrable standards.
These developments represented genuine progress because they reduced arbitrary authority, made professional standards more transparent and allowed expertise to travel beyond the individuals and institutions in which it had originally developed.
Yet something subtle happened in the process. The indicators through which competence could be recognised gradually became entangled with competence itself. Consistency came to signal reliability, procedural correctness to signal professionalism, measurable performance to signal effectiveness and confident communication to signal understanding. The difficulty begins when characteristics designed to make competence visible become the definition by which competence is understood.
The distinction matters because what can be measured and standardised tends to be what institutions become best at recognising. A professional who produces clear answers, follows established processes, communicates fluently, and performs consistently leaves behind abundant evidence of competence. Someone whose greatest contribution lies in recognising that the framework itself no longer fits the situation may leave considerably less. The more organisations depend upon observable signals to identify expertise, the greater the danger that they privilege what competence looks like over what competence actually does.
This transformation began long before AI, which is interesting precisely because it exposes a problem that technology did not create. Generative systems can now produce many of the signals that modern institutions have learned to associate with competence: articulate language, structured reasoning, rapid synthesis, polished presentations, confident recommendations and plausible explanations. What once required considerable professional effort can increasingly be produced within seconds.
The obvious conclusion is that AI threatens traditional forms of expertise. A more interesting possibility is that AI is revealing how impoverished our definition of expertise had already become. If eloquence can be generated, consistency automated, information retrieved instantly and plausible answers produced at negligible cost, then perhaps these qualities were never competence itself.
We may be entering an era in which the visible performance of competence becomes increasingly abundant while competence itself remains as difficult to cultivate as before. A beautifully structured analysis may reveal considerable judgment or merely competent use of a machine. The output alone increasingly tells us less about the intellectual process that produced it.
The useful response is not to retreat into some mysterious realm of human intuition that machines can never reach, but to ask which dimensions of competence become more important as its traditional signals become less informative. Here the older understanding becomes unexpectedly relevant, because it reminds us that competence was never merely the production of correct answers.
Competence also involves recognising which question deserves to be asked, distinguishing the relevant from the merely available, understanding when a general rule fits a particular situation and when it does not, identifying uncertainty rather than disguising it, and accepting responsibility for consequences that cannot be transferred to the system that informed the decision. None of these qualities requires rejecting AI. Intelligent systems may improve each of them by expanding the information available to professionals, challenging assumptions, and revealing patterns human beings would otherwise miss. But better information does not eliminate the need to decide what that information means in a particular context.
This matters for the way institutions educate, recruit and evaluate people. Universities may need to care less about whether students can reproduce convincing answers and more about whether they can interrogate them. Organisations may need to look beyond fluency and responsiveness when identifying talent. Professional education may need to make uncertainty, interpretation and responsibility more explicit. Leadership development may need to place greater value on recognising when an established framework no longer describes the reality it was designed to manage.
None of this requires abandoning standards, metrics or procedures. Modern institutions could not function without them, and a return to purely subjective assessments of expertise would recreate precisely the problems that standardisation was designed to solve. The challenge is to preserve the benefits of making competence visible without mistaking visibility for competence itself.
Perhaps, then, AI is not changing the meaning of competence so much as forcing us to notice that we had already changed it. Over decades, we moved from an understanding centred on judgment under uncertainty toward one increasingly associated with consistency, demonstrable performance and the reliable production of answers. That transition brought enormous benefits, but it also narrowed our field of vision. Now that machines can reproduce many of those visible characteristics with extraordinary ease, the distinction we allowed to fade has become difficult to ignore.
The question is therefore no longer whether AI will make human beings less competent. It is whether institutions will continue recognising competence through signals that technology has made increasingly unreliable, or whether we will recover a richer understanding of what competence was always supposed to describe: not simply knowing the answer, but knowing what to do when the answer is no longer enough.
