Site icon Critical Skills

Critical Skills: AI and the Future of Competence

Charles C. Jett

When Machines Can Do the Work — Judgment Becomes the Only Thing That Matters

This series is not a competition between frameworks. It is an examination of what professional competence actually requires — and which tools, used for the right purposes, best serve the professionals and organizations that depend on them.

Competency frameworks operate at two fundamentally different levels: systems designed for organizations to evaluate and develop leaders, and architectures individuals can use to guide their own professional growth.

The most important question any professional can ask about a framework is not whether it is comprehensive.

It is whether it was designed for them — and whether they can actually use it.

— • —

AI does not commoditize judgment. It commoditizes output.

That single distinction is what this article is about. And it is the distinction that every professional and every organization needs to understand before AI does for competence what social media did for expertise — make it impossible to tell the real thing from the performance of it.

For most of the modern professional era, competence and production were tightly coupled. If someone could produce sophisticated work, they almost certainly possessed the skills required to create it. A well-reasoned analysis revealed a capable analyst. A precisely argued brief revealed a practiced thinker. The work was evidence. It was harder to fake competence than to develop it.

Artificial intelligence has broken that coupling.

Production can now be generated without competence.

Polished reports, rigorous-looking analyses, fluent arguments, structured strategies — the visible outputs of expertise are now available to anyone with an internet connection and a prompt. The credential has been decoupled from the capability. The output has been decoupled from the judgment that was supposed to produce it.

That is the professional challenge of the next decade. And understanding it requires standing on the foundation this series has been building.

The first four articles in this series traced an unusual pattern in the research literature. Three independent frameworks — the McKinsey and Egon Zehnder leadership competency study, the Korn Ferry Leadership Architect built from Lominger’s sixty-seven-competency model, and the U.S. Department of Labor’s SCANS workforce framework — each converged on the same essential skill architecture without knowledge of each other, and without knowledge of the Eight Critical Skills that the executive search market had identified first. Corporate leadership research, enterprise talent management, and federal workforce policy each arrived, independently and afterward, at the same answer the market had already found.

When that happens, the result stops being a framework.

It becomes evidence.

That evidence is the Eight Critical Skills: Communication, Production, Information, Analysis, Interpersonal, Technology, Time Management, and Continuous Education.

This article is the payoff of that argument. It asks what happens when artificial intelligence can produce the visible outputs of those skills — and answers that the skills matter more than ever, not less.

But it also asks something harder: how do we find the real thing when the imitation is indistinguishable from it?

— • —

I. Judgment Is the Differentiator

Every major study of AI in the workplace arrives at the same boundary.

That boundary has a name. Economists call it Knightian uncertainty — the domain of risks that cannot be quantified because the probability distribution is genuinely unknown.

Frank Knight identified it in 1921, and a century later it remains the structural limit on what pattern-matching systems can do. When the situation is genuinely novel, algorithms have no prior patterns to match. Only humans can navigate it — and then only humans who have developed the judgment that comes from navigating real situations before.

What Judgment Actually Is

Gary Klein’s decades of research on naturalistic decision-making — built from studying firefighters, military commanders, and emergency physicians — reveals the mechanism. Expert judgment is not a deliberative comparison of options. Experts recognize patterns from accumulated experience, generate a course of action, mentally simulate it, and adapt — often within seconds. This is what Kahneman’s framework would call the refined product of System 2 thinking internalized into System 1: hard-won expertise operating as intuition. It is holistic situational awareness, not rule-following. It develops through practice. Real practice. Years of real decisions with real consequences.

And it is precisely what current AI architectures cannot replicate.

Philip Tetlock’s twenty-year study of expert forecasters provides the empirical measure of what separates good judgment from poor judgment. His superforecasters outperformed intelligence analysts with access to classified information by thirty percent. The qualities that distinguished them were not computational. They were metacognitive: intellectual humility, probabilistic thinking, willingness to update prior beliefs when evidence changed. Tetlock’s conclusion is direct: “Commitment to self-improvement is the strongest predictor of performance.” These are quintessentially human capacities. And they are precisely the capacities that AI use, without deliberate cultivation, tends to erode.

Practical Wisdom Cannot Be Outsourced

Aristotle called it phronesis — practical wisdom, the capacity to choose the right action in the right circumstances for the right reasons.

Research on AI and phronesis identifies the risk most commentary misses: if AI absorbs an increasing share of the decisions a professional makes, the development of judgment slows. The professional produces better outputs in the short term. Their underlying capability atrophies in the long term.

Phronesis does not develop by reading about good decisions. It develops by making them — including making wrong ones, living with the consequences, and adjusting.

That process is not a side effect of professional practice. It is the mechanism by which the Eight Critical Skills compound into judgment over time. Communication skill practiced under real pressure becomes the judgment to know when silence is the right answer. Analysis skill tested against consequential problems becomes the judgment to know when the model is wrong. Every skill in the framework is, in the end, a pathway to phronesis — if it is actually practiced.

“The people who will use AI best are the ones who already know their domain deeply.” Judgment is not a soft skill. It is the hard skill that all other skills depend on.

What the Leading Thinkers Have Found

The researchers who have studied this most carefully reach a consistent conclusion.

Three researchers. Three disciplines.

One finding: Judgment is not a gap that more computing power will close. It is a structural distinction between human cognition and algorithmic pattern-matching.

— • —

II. Skills Are More Important Than Ever

There is a seductive logic in the idea that AI makes skills less important. If AI can write the report, analyze the data, and draft the code — why invest in the underlying capabilities?

Because the prompt is only as good as the judgment behind it. And the judgment is only as good as the knowledge that frames it. AI democratizes production. It does not democratize competence.

The Jagged Frontier: What the Data Actually Shows

The most important empirical study of AI in professional work is Dell’Acqua, Mollick, and colleagues’ field experiment at Boston Consulting Group.

Seven hundred fifty-eight BCG consultants were tested on realistic business tasks. For tasks inside AI’s capability boundary, consultants using AI completed twelve percent more tasks, worked twenty-five percent faster, and produced outputs rated more than forty percent higher in quality. Lower-performing consultants improved forty-three percent. Higher-performing consultants improved seventeen percent. AI compressed the distribution — raising the floor faster than it raised the ceiling.

But the finding that matters most is the one buried beneath the headline.

For tasks outside AI’s capability boundary — novel situations requiring genuine domain judgment — consultants using AI were nineteen percentage points less likely to produce correct solutions. Not because the AI was obviously wrong. Because it generated confident, coherent, plausible answers that happened to be incorrect. And the consultants trusted them.

Mollick named this the “jagged technological frontier”:

AI capabilities are uneven, unpredictable, and — critically — invisible from the outside.

You cannot see the boundary from outside it. Knowing when to trust AI and when to override it is itself a skill. And it is a skill that only genuine domain expertise provides.

The Floor Rises. The Ceiling Holds.

Brynjolfsson, Li, and Raymond’s study of 5,172 customer support agents — published in the Quarterly Journal of Economics in 2025 — found that AI increased average productivity fifteen percent. Novice and low-skilled workers improved thirty-four to thirty-five percent. The floor rose dramatically. The ceiling held.

The data reveals why: the AI system learned from top performers.

Their tacit knowledge — the accumulated expertise of their best decisions, their best resolutions — became the training substrate. Skilled workers became more valuable as the source of institutional intelligence that AI then distributed across the organization. AI amplified their expertise. It did not replace it. And Noy and Zhang’s companion study found the same structural shift: AI absorbed production, pushing professional work toward idea generation, critical evaluation, and judgment. Higher-order skills became the actual job.

What the Institutions Are Saying

The World Economic Forum’s Future of Jobs Report 2025 — based on surveys of over one thousand employers across fifty-five economies — projects 170 million new jobs created and 92 million displaced by 2030.

Thirty-nine percent of workers’ core skills will change by that date. The top growing skills: analytical thinking, creative thinking, resilience, curiosity, and technological literacy.

Every one of those skills maps directly onto the Eight Critical Skills framework. The market has not changed what it needs. It has raised the stakes for having it.

McKinsey’s State of AI 2025 survey found that eighty-eight percent of organizations now use AI in at least one function. Seventy-nine percent use generative AI specifically. But only six percent qualify as AI high performers — organizations actually capturing significant enterprise value.

The gap between adoption and mastery is the skills gap.

David Autor at MIT frames the structural argument: “The unique opportunity AI offers is to extend the relevance, reach, and value of human expertise.”

AI does not eliminate the need for expertise. It changes how expertise is deployed — and makes the professionals who genuinely have it more valuable, not less.

— • —

III. AI Amplifies the Appearance of Competence

Section I established what judgment is and why AI cannot replicate it. This section addresses the internal challenge that creates for individual professionals: AI can make you appear more competent than you are — and you may not know it is happening.

In every prior era, the outputs of competence were hard to produce without the underlying capability. A poorly reasoned analysis revealed a poorly developed analyst. The work was evidence. It was harder to fake competence than to develop it. That has changed.

The Evidence Is Unambiguous

Expert evaluators rated ChatGPT-generated essays higher in quality than human-written ones — performing significantly better on logic and fluency (Herbold et al., 2023, Scientific Reports).

A 2024 study found that non-expert readers performed below chance at identifying AI-generated poetry — they were actually more likely to judge AI poems as human-authored. A 2025 study found that humans correctly identified professional-level AI-generated text less than twenty percent of the time.

The professional exam results are equally stark. GPT-4 scored at the ninetieth percentile on the Uniform Bar Exam, exceeding the passing threshold for all US jurisdictions. It scored eighty-six percent on the USMLE Step 1 medical licensing exam. A meta-analysis across multiple licensing exams found GPT-4 correctly answering approximately seventy-five point nine percent of questions.

The credential — always an imperfect proxy for competence — has now been decoupled from it at scale.

The Dunning-Kruger Effect Inverts

The most consequential study in this space was published in 2026 by Fernandes, Welsch, and colleagues at Aalto University: “AI Makes You Smarter but None the Wiser.” In two large-scale studies using LSAT logical reasoning tasks, participants using AI improved their performance by three points. They overestimated their performance by four.

The classic Dunning-Kruger effect — where incompetent people overestimate their ability because they lack the metacognitive capacity to recognize incompetence — did not just persist. It reversed. Everyone overestimated, regardless of skill level.

The most technically sophisticated AI users were the most overconfident.

Professor Robin Welsch observed: “When it comes to AI, the DKE vanishes. Higher AI literacy brings more overconfidence.” The behavioral explanation is direct: most users prompted AI once, accepted the output, and moved on. One interaction. No verification. No probing. The competence illusion is not theoretical. It is measurable, it is spreading, and it is invisible to the people experiencing it.

This is also where the professional character foundation matters. The SCANS framework explicitly included personal qualities — integrity, responsibility, self-management — as baseline requirements that no skill set replaces. In an era where AI makes it easy to produce work that looks competent, the internal discipline to ask whether it actually is competent is not a soft consideration. It is the professional standard. The Dunning-Kruger inversion is not just a cognitive phenomenon. It is an ethical one.

Cognitive Offloading and the Deskilling Spiral

Sparrow, Liu, and Wegner established the “Google effect” in 2011 — when people expect information to remain available externally, they show lower recall of the information itself.

The AI version of this is more severe.

Gerlich (2025) found a negative correlation between frequent AI use and critical-thinking abilities. A 2025 CHI study documented practitioners reporting erosion of core professional skills through over-reliance.

A 2026 paper in Frontiers in Medicine introduces a concept that may be the most alarming in this literature: never-skilling.

“When the learning opportunities that develop adaptive expertise are systematically reduced, judgment, flexibility, and retention of mechanistic understanding weaken.” The professional who uses AI from day one may never build the foundation that makes AI use genuinely productive.

The competence illusion is not theoretical. It is measurable, it is spreading, and it is invisible to the people experiencing it.

The Kenyan Entrepreneurs — AI as Amplifier, Not Equalizer

The HBS field study with 640 entrepreneurs in Kenya given access to a GPT-4-based business coach produced results that crystallize the internal challenge.

High-performing entrepreneurs saw profit gains of ten to fifteen percent. Struggling entrepreneurs did worse than the control group.

The difference was entirely attributable to judgment.

AI amplified the gap it was designed to close.

The implication is direct: AI does not create competence. It extends it.

The professionals who will benefit most are not the ones who need AI most.

They are the ones who need it least.

— • —

IV. Real Competence Becomes Harder to Detect

The internal challenge — cognitive offloading, the competence illusion, the deskilling spiral — is what happens inside the individual professional. This section addresses the external challenge: what it means for the organizations trying to identify, hire, and develop real talent in an environment where the imitation has become almost indistinguishable from the real thing.

The AI Arms Race in Hiring Has Made Everything Worse

SHRM reports that forty to eighty percent of job applicants now use AI to write resumes, craft cover letters, and prepare for interviews. Sixty-four percent of recruiters reported an increase in look-alike AI-generated resumes — which paradoxically increased screening workload.

Nichol Bradford of SHRM states the problem plainly:

“The AI arms race does not benefit either side. Recruiters cannot go through thousands of applications. Job seekers are demoralized to never hear from a human.”

SHRM’s 2025 Benchmarking Survey shows that average cost-per-hire and time-to-hire have both increased over the past three years — the exact period of accelerating AI adoption.

The confidence problem compounds it. Eighty-eight percent of hiring managers report they can detect AI-assisted applications. Only thirty-three percent can spot an AI-generated resume in under twenty seconds. For professional-level AI text, detection rates fall below twenty percent. Evaluators are overconfident about their ability to detect exactly what they are missing — their own version of the Dunning-Kruger inversion documented in Section III.

What Reliable Signals Look Like Now

Some signals have held their validity. Others have collapsed.

Structured behavioral interviews remain the gold standard.

Sackett et al.’s 2022 meta-analysis found a validity coefficient of 0.42 — the highest of any assessment method. These interviews require candidates to draw on real experience with specific, probeable details — something AI-coached preparation cannot manufacture under live, adaptive follow-up. The tacit knowledge that Klein’s research describes manifests under this kind of pressure. It cannot be faked in real time.

Laszlo Bock, who ran People Operations at Google during its most rigorous talent development period, reached an identical conclusion independently. Google’s number one hiring criterion was learning ability — not credentials, not prior expertise. “The ability to process on the fly. The ability to pull together disparate bits of information.” Google eliminated brain teasers as “a complete waste of time” and validated structured behavioral interviews as the most reliable predictor.

Credentials were ranked the least predictive attribute after the first two career years.

Michael Polanyi’s concept of tacit knowledge — “we can know more than we can tell” — provides the theoretical grounding.

Tacit knowledge manifests in practice and deviates from formal procedure.

It is what distinguishes the expert surgeon from the competent one, the experienced manager from the credentialed one. It develops through deliberate practice over years. AI can read Polanyi. It cannot develop tacit knowledge. And it cannot convincingly simulate it under real-time professional pressure.

Organizations Are Adding Friction Back

The organizational response is instructive. Gartner reports that companies which created hyper-efficient application processes are now deliberately adding friction — longer applications, live problem-solving tasks, and proctored skills assessments that AI cannot take on a candidate’s behalf.

Gartner projects that by 2027, seventy-five percent of hiring processes will include certifications and tests for workplace AI proficiency — not to demonstrate AI use, but to demonstrate that the professional can evaluate and direct it.

The shift toward skills-based hiring is accelerating. LinkedIn’s 2025 data shows companies with the most skills-based approaches are twelve percent more likely to make quality hires.

Todd Rose’s research established the mathematical case against standardized proxies before AI made the argument urgent. When 4,063 Air Force pilots were measured on ten body dimensions, zero were average on all ten. A cockpit designed for the average pilot fit no actual pilot. Standardized screening processes — including AI-augmented ones — systematically miss the exceptional because the exceptional is, by definition, non-average.

— • —

V. What It Means for the Eight Critical Skills

The prior articles established the Eight Critical Skills through convergence — three independent research efforts arriving at the same architecture.

This article tests that architecture against a fundamentally different environment. The finding: the Eight Critical Skills do not become less important in the AI era. They become the differentiating factor in it.

Every skill in the framework now operates at a higher cognitive level — because AI has absorbed the lower-level functions and pushed human work upward. And every skill is now, more explicitly than before, a pathway to the judgment that Section I identified as the ultimate differentiator. The connection is direct: mastering the Eight Critical Skills over time is how phronesis develops.

The skills are not ends in themselves. They are the practice through which judgment is built.

The Eight Critical Skills are not a relic of a pre-AI world.

They are the map to navigating the AI world.

And the map leads, ultimately, to the same destination it always has: the practiced professional whose judgment cannot be automated, whose expertise cannot be faked, and whose contribution cannot be replaced by a prompt.

— • —

The Finding

Five articles. Four independent research frameworks — McKinsey and Egon Zehnder, Korn Ferry and Lominger, SCANS, and the frontier of AI research itself. One convergent finding:

Artificial intelligence has not changed what professional competence requires. It has exposed it.

For decades, the Eight Critical Skills described the capabilities that produced good work. In the AI era, they describe the capabilities required to know whether the work is good at all.

Before AI, competence produced output.

Now output can be generated without competence — which means the only remaining test of competence is judgment: the capacity to evaluate, direct, question, and decide that AI cannot supply.

The competence illusion is real, measurable, and spreading. AI enables people to produce polished outputs without the underlying capability. Hiring processes are struggling to detect the difference. Assessment methods built for a pre-AI world are failing. The Fernandes et al. finding — that AI use causes universal overconfidence and eliminates the Dunning-Kruger correction — is not a warning about other people. It is a warning about every professional who uses AI without the metacognitive discipline to question it.

AI makes production easy. Judgment remains hard. The professionals who build it — through real practice, real decisions, real consequences — will lead the AI era.

The ones who do not will merely prompt it.

AI does not commoditize judgment. It commoditizes output. And in doing so, it makes judgment — real judgment, developed through real experience — the rarest and most valuable professional asset in the room.

The series that began with McKinsey and Egon Zehnder’s work on leadership competencies, moved through Korn Ferry’s sixty-seven-competency architecture, traced the convergence with SCANS, and arrives here — in an AI era that has changed the operating environment without changing what the market has always needed.

Eight skills.

The same eight skills.

More important than ever.

— • —

About This Series

The Critical Skills Series examines what professional competence actually requires — through the lens of the research that has tried hardest to define it. Part One introduced the Eight Critical Skills framework. Parts Two and Three tested it against McKinsey/Egon Zehnder and Korn Ferry/Lominger. Part Four established convergent validity with the federal SCANS framework. This article is Part Five.

 

Selected Research Citations

 

Copyright © 2026 by Charles Cranston Jett

Exit mobile version