Last month I was on a stage at the Athens College Theater, at the MIT Universal AI Summit, alongside my old friend George Doukidis of the Athens University of Economics and Business and Corallia Ventures. Our moderator, Symeon Retalis of the University of Piraeus, opened with the question the whole conference had been circling: what separates AI reskilling that actually changes how people work from a course catalog no one finishes?

My answer began with a confession that a business school professor probably should not make in public. The more I study firm-level AI reskilling, the more convinced I become that it does not belong in a school at all. Not in mine, not in anyone’s. Because the biggest change AI brings to organizations is not a new tool or even a new skill set. It is the end of the separation between working and learning.

A recent invention

We treat that separation as a law of nature. It is not. It is a recent invention. For most of human history, working and learning were the same activity. You learned the trade by doing the trade, next to someone who already knew it. The Industrial Revolution pulled them apart: now you work, now you learn. We built classrooms on that split, then curricula, then degrees, and eventually the corporate training department, the course catalog, and the completion certificate.

The split made sense when skills lasted a career. It stops making sense when skills turn over every six to twelve months. If you need a capability now, you cannot wait for the course to be designed, approved, scheduled, and delivered. By the time it arrives, the skill it teaches may already be obsolete. Learning has to move back inside the work, to the moment the gap appears. AI is what finally makes that possible at scale, which is why I believe it will not just change what we learn but dissolve the two-century-old boundary around where and when we learn it.

Courses are the 10 percent

This is why the standard corporate response to AI, buying a course catalog, misses the point so completely. In the United States, it is becoming common wisdom that in any serious reskilling effort, courses are maybe 10 percent. The intuition is not new: the Center for Creative Leadership’s 70-20-10 framework has held for four decades that formal coursework accounts for roughly a tenth of how people actually develop, with the rest coming from challenging assignments and developmental relationships. What is new is that AI-era skill turnover makes ignoring this finding fatal. The other 90 percent is the learning system you construct around the work itself: stretch assignments that let people practice new capabilities on real problems, peer mentoring and mentoring of every other kind, and continuous assessment rather than a final exam and a certificate.

Notice what that list is. It is not a curriculum. It is an organizational design. Which is precisely why it cannot be outsourced to a school, mine included. A university can hand you the 10 percent. Only you can build the 90.

You cannot hit a target you cannot name

But the 90 percent has a prerequisite that most organizations fail before they start: knowing what the target is. Ask a company what skills it wants its people to develop, and you will get a list of nouns. Leadership. Critical thinking. Collaboration. Ask ten people in that same company what those nouns mean, and you will get ten different answers. A noun cannot be measured, cannot be assessed, and cannot be developed on purpose. If your competency model is a list of nouns, you do not have a competency model. You have a mood board.

This is the argument I made in Hire for Verbs, Not Nouns, and it matters even more for reskilling than for hiring: what looks like a talent problem is usually a language problem. The alternative is to define every skill as observable behavior in context. Not “leadership,” but “the ability to define a mission and persuade a team to follow it.” Not “communication,” but “the ability to turn a difficult conversation around.” This is unglamorous, contentious, thoroughly human work. Nobody can automate it for you, because it amounts to deciding what your organization actually values. But it is the keystone. Every other piece of the system depends on it.

Then the machines can help

Here is what surprises people: once the definitions exist, the technology is essentially ready. Imagine configuring your AI assistant, the Copilot you already have, to observe your work against the competencies you have chosen to develop. Your emails, your Slack threads, your meetings. Then it gives you feedback, daily or weekly, as you prefer: here is where you did this well, here is a moment you could have handled differently, and here is a fifteen-minute resource for the gap that keeps showing up, delivered now, not next semester. This is not science fiction. I know executives in the United States who are doing versions of it today.

Take it one step further, with governance designed carefully so that the coach does not become Big Brother, and the same evidence can feed evaluation. Instead of the annual narrative review, which research has repeatedly shown to be biased and imprecise, you assess people on vignettes: documented moments of actual performance, with the employee controlling which ones enter the record. The moment you mentored a colleague effectively. The moment you turned a difficult conversation around. This is the shift I described in Gen AI Could Fix Performance Reviews—or Make Them Even Worse: the danger is using AI to write ever more persuasive narratives about performance, and the opportunity is using it to surface evidence of what people actually did.

Longtime readers will recognize the destination, because it is the through-line of this newsletter. In Why Résumés Lie, I argued that the résumé has always promised ability and delivered biography, with generative AI merely the final straw, and in The Movie of Work Is Already Playing I showed that aviation and medicine already certify people on demonstrated performance rather than paper. A dynamic, verifiable record of performance vignettes is all of these ideas converging. It starts inside companies, as a better performance review. It ends as the résumé of the future, and eventually the transcript: why shouldn’t universities send graduates into the world with a living digital passport of evidenced abilities rather than a static list of grades that correlate weakly with anything?

George, incidentally, supplied the proof that fusing work, learning, and certification is commercially viable, and it came with a dose of national pride: PeopleCert, Greece’s first unicorn and the owner of PRINCE2 and ITIL. They do not just sell exams. They embed a methodology into the organization; people learn by working inside it, and the certificate attests to that lived practice. Assessment and work, fused. The thesis of this essay, arriving from the opposite shore.

The competency checkup

If working and learning become one continuum, something else follows, and it is the idea from the panel I keep returning to. Skills now rise and fall like public health statistics, and companies like Lightcast already track and forecast that movement. So picture this: twice a year, the way you get a medical checkup, you get a competency checkup. A system examines your skill profile against where the labor market is heading and tells you which of your skills will be obsolete in three years, which adjacent skills are rising, and what to do about it. Blood test, diagnosis, prescription. For your career. The systems do not quite exist yet. Every component does.

The one course I would still build

After all this, you might conclude I think business schools have no role left. Not quite. There is one course I would build tomorrow, and it was my closing plea in Athens, aimed at the Greek universities in the room: educate business leaders toward a skills mindset. Teach them to define skills as observable behavior, to assess them with evidence, and to make them the first-order object around which they organize their business, in place of credentials, résumés, and prestige.

Because the labor market, by and large, is not there yet. Most organizations still hire on degrees and pedigree even as the signal decays; the discounting of AI-exposed master’s degrees I documented in AI Came for the Master’s Degree shows the correction beginning, but it is early and uneven. The technology to coach, assess, and credential people continuously already exists. The bottleneck is the mindset.

The training room served us well for two centuries. The era it belonged to is ending. Working and learning are becoming one thing again, the way they were for most of human history, and the organizations that internalize this first will not just reskill faster. They will finally know what their people can actually do.

What would a competency checkup reveal about your skill set? Reply in the comments or email me at dell@bu.edu

American universities have never awarded more master’s degrees than they are awarding right now. The class of 2025 collected roughly 949,000 of them, up from 815,000 in 2018. That is growth in every single year, straight through the arrival of ChatGPT.

Which is what makes the other side of the ledger so strange. In a new working paper with my Boston University colleagues Patricia Cortés and Qi Wang, we examined more than 100 million U.S. job transitions between 2018 and 2025 and traced what employers actually did at the hiring desk. In the occupations most exposed to generative AI, something flipped after late 2022. For years, employers there had picked master’s holders a bit more often than their numbers among available candidates would suggest. After ChatGPT, they began picking them less often — even as the candidate pool filled with more of them. All told, master’s hiring in the most AI-exposed occupations fell by roughly 10 percent relative to the least exposed ones, a gap that opens right after ChatGPT’s release and widens every year since. The graduates kept coming. The employers quietly started passing.

But the strangest part is whose résumés the shift shows up on. The decline is concentrated entirely among candidates who already had work experience. For people entering the labor market for the first time, the master’s degree carried exactly as much weight as before. The market didn’t stop valuing the credential across the board. It stopped consulting it whenever it had something better to read.

That pattern is not a fluke. It’s a clue. And once you see what it’s a clue to, it changes how you should think about what a degree actually is.

How we measured it

We used data from Revelio Labs, which reconstructs employment histories from public professional profiles. That let us observe not what employers say they require in job postings, but who they actually hire. We focused on entry-level positions in occupations where a master’s degree is a genuine choice rather than a legal requirement: accountants, market research analysts, business intelligence analysts, paralegals, writers.

To decide which occupations count as AI-exposed, we used the most widely cited measure in the field: the “GPTs are GPTs” index, built by a team of researchers from OpenAI and the University of Pennsylvania. The idea is simple. Take the government’s catalog of what workers in each occupation actually do, and ask of every task: could an AI assistant help a worker do this at least twice as fast, just as well? Add up the answers, and every occupation gets a score, measured the day the technology arrived, before employers had any time to react.

Then we compared the most exposed fifth of occupations against the least exposed fifth, year by year. For four straight years before 2022, the two groups moved in lockstep. Then ChatGPT arrived, and the lines split apart, slowly at first, then faster.

Three more facts sharpen the picture. The decline is concentrated entirely in non-STEM degrees, especially business master’s degrees: the programs that teach synthesis, analysis, forecasting, and professional writing, which is to say, the tasks ChatGPT is best at. STEM master’s hiring didn’t budge. The decline is strongest at exactly the firms where Census data says AI is actually being deployed. And it can’t be a shortage of graduates: as we saw, the supply grew every single year. Employers didn’t run out of credentialed candidates. They started passing on them.

One credential, two signals

So why would AI make a master’s degree count for less when the candidate is arriving from a similar job at another firm, but not when she’s a 24-year-old with a blank résumé?

Because a degree was never one thing. It’s a bundle.

Part of what a master’s degree tells an employer is this person has specific, advanced skills: she can build the forecast, structure the analysis, write the report. Call that the certification component. The other part is more primal: this person got into a demanding program, survived it, and finished it. That signals intelligence, persistence, follow-through. General ability. Call that the signaling component.

For fifty years, ever since the economist Michael Spence formalized the idea, we’ve treated the two as welded together. Generative AI is the crowbar that pries them apart. When AI can produce a competent first draft of the analysis, the forecast, and the report, the certification component loses value: the skills are still real, just no longer scarce. The signaling component is untouched. Arguably it matters more than ever.

Now the puzzle resolves itself. Evaluating a first-time job seeker, an employer has almost nothing else to go on, so the degree keeps its full weight. Evaluating an experienced candidate, the employer has something better: a work history, a record of actual accomplishments, what I’ve been calling the “movie of work.” Once that alternative exists, the degree is valued mainly for its skill certification, precisely the component AI just devalued. So its weight in the hiring decision falls.

The degree didn’t die. It got unbundled. And the market now prices the two components separately.

What this means if you run a graduate program

Resist two tempting misreadings of this evidence.

The first is denial: “enrollments are fine, nothing to see here.” True, for now. But employer behavior leads student behavior by a few years. When working professionals notice that the MBA or the MS in analytics no longer moves the needle with hiring managers, demand will follow. And they will notice; they are the most ROI-sensitive customers in higher education.

The second is fatalism: “AI killed the master’s degree.” It didn’t. It killed a particular value proposition: paying six figures to certify skills a $20-a-month subscription can now approximate. The way forward is to rebuild curricula around competencies that complement AI rather than compete with it.

Take the MBA. Its traditional core of financial modeling, forecasting, market analysis, and professional writing is precisely the territory AI has claimed. A reinvented MBA would certify what AI can’t.

• Problem framing: AI is remarkable at answering questions and useless at deciding which question matters.

• Judgment: choosing among several plausible AI-generated analyses when the data doesn’t settle it, and owning the consequences.

• Artifact reasoning: interrogating a machine-generated model or memo, sensing where it’s subtly wrong, taking responsibility for what goes out the door.

• Leadership and persuasion: moving actual humans remains stubbornly hard to automate.

• Orchestration: designing work so that people and AI each do what they’re best at.

None of these is soft. All are teachable and, crucially, assessable, provided programs are willing to grade performance rather than polished deliverables.

The unbundling also argues for unbundled products: modular credentials, stackable certificates, and continuous reskilling relationships that let learners buy the components that still carry value and refresh them as the technology moves.

The deeper lesson is this. Universities have long treated the informational value of their credentials as a constant, an endowment like the campus or the brand. Our results say it’s a variable. It moves when technology moves.

What this means if you’re deciding whether to get one

The calculus now splits by career stage.

If you’re just starting out, the degree’s signaling value is intact: employers still lean on it when they have nothing else to go on. But understand what you’re buying, which is proof of general ability, less and less a skills advantage. And ask whether there’s a cheaper way to prove it.

If you’re a few years in and eyeing a business or analytics master’s to power your next move, pause. You are exactly the person for whom the credential is losing value fastest, because you already have the thing employers now trust more: a track record. The better investment is making that record visible through portfolios, shipped projects, and verifiable outcomes, and building the capabilities no transcript captures.

The master’s degree spent a century as a bundle of skills and signal, sold at a single price. AI has broken the bundle open. What happens next depends on whether institutions reprice the parts — or keep charging for a package the market has already taken apart.

The working paper, “Generative AI and the Informational Value of Educational Credentials: Evidence from the Master’s Margin” (with Patricia Cortes and Qi Wang), is available on SSRN.

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7045118

One of the most surprising things about AI is that it is forcing us to confront problems we have quietly lived with for years.

Take performance reviews.

For decades, organizations have relied on managers to write narratives about employee performance. We know those narratives are often shaped by memory, storytelling ability, personal relationships, and countless other sources of bias. Yet because writing good reviews was difficult, the limitations of the system were partly hidden.

Now, generative AI can produce polished evaluations in seconds.

At first glance, this seems like progress. But there is a danger. When AI makes every review sound thoughtful and insightful, it may become even harder to distinguish careful evaluation from weak evidence.

The real opportunity lies elsewhere. Rather than using AI to write better stories about performance, we can use it to surface better evidence of performance: the decisions people made, the problems they solved, the colleagues they influenced, and the moments where judgment mattered.

We settled for narratives not because they were ideal, but because the alternative was too expensive. The evidence was buried in emails, documents, meeting notes, and project records. AI changes that. For the first time, organizations can uncover and organize performance evidence at scale.

In my latest Harvard Business Review article, I argue that AI could help shift performance reviews from narratives about work to direct evidence of work itself.

Read the article here: [HBR link]

I’d be interested to hear whether you think AI will make performance evaluations more trustworthy—or simply more persuasive.

#AI #Leadership #Management #PerformanceReviews #FutureOfWork #HumanCapital

Every year, I teach a course on databases.

For weeks, I show students how to build data models and write SQL queries—carefully, step by step.

This year, during one of my lectures, a student typed a few lines of the case into Claude. Seconds later, he had a clean, well-structured data model—something I normally spend weeks teaching.

For a moment, it felt like the course had collapsed. Then came the real problem:

No one in the room could tell if the answer was right.

That’s when the question changed.

Not “How do we stop students from using AI?”
But “What are we actually trying to teach?”

I explore that question in this new piece in The Chronicle of Higher Education:

👉 https://www.chronicle.com/article/when-ai-can-do-everything-what-is-left-to-learn

In short: if AI can produce the artifact, our courses must teach students to interrogate it—frame the problem, question the output, and take responsibility for the answer.

For those of you who do not have access to The Chronicle, here is the full text of my article:

Last semester, a student stopped me in mid-lecture. We were discussing how an online retailer should organize its customer and order data when he typed a few sentences into his laptop and turned the screen toward me. ChatGPT had produced a clean data model — customers, orders, products, payments — that looked remarkably like the diagrams I normally spend weeks teaching students to construct themselves.

For a moment, my entire course felt strangely obsolete. But only for a moment. Then, I realized something even more troubling: nobody in the room — neither the student who had generated it, nor any of his classmates — could say whether the AI-generated model was correct.

I suspect many faculty have had a version of this moment. The most common response is to treat it as an assessment problem: How do we stop students from submitting AI-generated work? How do we design assignments AI can’t complete?

These are real questions. But they start in the wrong place. The deeper question is simpler and more fundamental: what do we want students to be able to do in a world of intelligent machines, and do our courses still develop those capabilities?

My answer, after redesigning a course around this question, is that generative AI has forced a reckoning not just with our assessments but with our learning outcomes. The faculty who engage that reckoning seriously will find that it leads to somewhere more interesting than a revision of academic integrity policies.

*

For decades, higher education relied on a convenient shortcut: if students could produce a certain type of artifact, we assumed they had developed the competency behind it. Writing an essay meant that a student could construct an argument. Writing code meant they understood computation. Producing an analysis meant they could reason about data. This worked because producing the artifact required mastering the competency.

But the creation of artifacts was never really the goal. It was the exercise, the mechanism by which students developed the mental models we really cared about. The essay forced students to structure an argument. The program forced them to trace the logic of an algorithm. The analysis forced them to specify a question precisely enough that a given data set could answer it. We were building cognitive capabilities. The artifact was how we made that cognitive development visible.

Generative AI has separated those two things. Students can now produce many of the artifacts without developing the capabilities that once came with them. The exercise no longer reliably produces the outcome. Which means that we need to redesign the exercise, not just the way we assess it.

*

The answer begins with distinguishing between two kinds of skills that AI has pulled apart. The first is artifact production: generating the essay, the program, the diagram, the analysis, the competency most of our learning outcomes have been built to develop, and the one generative AI has made widely accessible regardless of whether genuine understanding accompanies it.

The second is artifact reasoning: deciding what artifact should exist, directing a system toward producing it correctly, and judging whether the result answers the question it was meant to address. This is what AI cannot do. It is also what graduates are increasingly expected to do in the workplace: supervise, evaluate, and take responsibility for outputs that intelligent systems generate on their behalf.

Artifact reasoning, in turn, has two components. The first is problem framing: specifying a question precisely enough that an artifact can answer it. Consider a question deans often ask: Which courses are most effective? Before any algorithm can answer it, someone must decide what “effective” means. Higher exam scores? Greater improvement between the first and last assignment? Better performance in follow-on courses? Or something harder to measure, like the ability to apply ideas in unfamiliar situations? Each definition leads to a different data set, a different model, and ultimately a different answer. An AI system can generate dashboards and rankings once the question is posed. But unless someone frames the problem carefully, the system will produce precise answers to a vaguely defined question.

The second component of artifact reasoning is critical interpretation: reading an artifact carefully enough to know whether it answers the intended question, what assumptions it encodes, and where it might mislead. Recognizing these failures requires genuine comprehension of the underlying system, the kind that only develops through sustained engagement with the material.

Taken together, these constitute the new core capability our courses need to develop, not as an add-on to existing learning outcomes but as the organizing framework around which courses should be redesigned.

*

My database course at Boston University’s Questrom School of Business was where I worked this out. The course covers tools that are central to how organizations operate: data models, database queries, analytical dashboards. What changed was the question each course unit was designed to answer. In earlier versions: Can students produce these artifacts correctly? In the redesigned version: Can students reason about what these artifacts do?

Take data modeling, the first topic students encounter. Traditionally, students read a description of a company’s operations and drew a diagram of its key entities and relationships. The exercise ended when the diagram looked correct. Now, the class begins with a deceptively simple question: What exactly are we trying to represent?

What counts as a customer? Should payments be modeled separately from orders? What happens when a single order ships from multiple warehouses? Before drawing anything, the class must decide what aspects of a business’s messy reality should be captured as structured data, and what each choice will mean for every analysis the database will ever support. Only then do students construct models, comparing their versions with AI-generated alternatives. The interesting part comes when the diagrams differ. Each looks plausible. Each encodes different assumptions. The exercise is no longer about producing a correct diagram. It is about explaining what each diagram means.

The same shift runs through the course’s treatment of database queries. Students begin not by writing code but by clarifying the underlying business question itself. Suppose a company asks which customers generated the most revenue last month. Before writing a database query to determine the answer, the class must decide what the question really means. What counts as revenue? Does “last month” refer to orders placed or orders delivered? How should returns be handled? Only after specifying the logic do students generate the query, sometimes themselves, sometimes with AI. Their task then becomes explaining what the query actually computes. Two queries can run without errors yet answer subtly different questions. Only one answers the question that was intended.

Later in the semester, students build dashboards. But even here, the exercise begins differently: What decisions should this dashboard inform? Before creating any visualization, students must explain what questions it is meant to answer. A dashboard, they discover, is not merely a visual display. It is an argument about what the data means.

Throughout, students are encouraged to use AI tools to generate models, queries, and visualizations. But every assignment requires them to explain what the artifact does, what assumptions it relies on, and how it answers — or fails to answer — the original question. Each module begins with examples in which AI-generated outputs contain subtle errors or misleading assumptions. Once students understand that they cannot reliably evaluate AI output without gaining hands-on experience on what correct output looks like, the temptation to take shortcuts largely disappears.

By the end of the semester, class discussions had become more animated than at any point in my years teaching the course. Students are not asking how to produce the next artifact. They are asking why a design choice matters, what an analysis is hiding, and how a different framing would change the result.

*

The shift toward artifact reasoning does not mean abandoning production. Some hands-on production experience is indispensable. Students who have never written a query themselves have a genuinely difficult time evaluating queries written by AI. Without having constructed a data model from scratch, they struggle to see why two plausible-looking models might encode meaningfully different views of a business. Production is not the destination, it is the road — and there is no shortcut around it.

Students need enough production experience to be able to think with the artifacts they encounter. They don’t need to become world experts. What they need is the kind of familiarity that comes from having struggled with the construction process enough times to understand what “good” looks like, and what decisions it requires to be generated correctly.

A useful analogy: a music listener who has studied an instrument, even at an amateur level, hears differently from one who has not. They notice the difficulty of a passage. They understand what the performer chose to emphasize and what that choice cost. They can evaluate an interpretation rather than merely receive it. They do not need to perform at a professional level to reason and make judgments about what they hear. But they do need enough production experience to develop a mental model of how the music is made.

In practice, this means production exercises should remain in our courses, but we should always think of them as means to an end. The goal of having students write a query is no longer to develop query-writing fluency as an end in itself. It is to build the mental model that reasoning subsequently requires. The debrief after each production exercise matters as much as the exercise itself: not “did it work?” but “what did you have to decide to make it work, and what would have happened if you had decided differently?”

*

Every discipline that has relied on artifact production as the primary evidence of learning now faces the same reckoning.

In writing courses, the outcome that matters now is not producing a well-structured argument —AI can do that — but analyzing one: identifying where evidence is weak, explaining how conclusions depend on assumptions that might not hold, and knowing how to make improvements.

In computer science and data science, the shift is from producing correct code to identifying edge cases, understanding what a model assumes, and recognizing when an algorithm that runs is not an algorithm that works for the intended purpose.

In graphic design, the emphasis moves from generating images to judging them: articulating why a design communicates effectively, recognizing when visual choices contradict the intended message, and refining the result so that form and purpose align.

In each case, the production competency does not disappear. It becomes a prerequisite to the reasoning competency, the foundation that makes critical interpretation possible. What changes is which one we treat as the destination.

*

This is a daunting task, and professors have every incentive to evade it. It requires revisiting learning outcomes that have been stable for years and accepting that some of what we have long measured is no longer a reliable proxy for the capabilities we care about.

There is an argument, sometimes made in faculty meetings, that we should restrict AI use and preserve our existing learning outcomes. The deeper problem with this argument is that it prepares students for a world that no longer exists. The graduates we send into the workforce will spend their careers working alongside AI systems, responsible for the outputs those systems generate. A curriculum that withholds AI, or that continues treating production as the primary evidence of competency, is not protecting rigor. It is graduating students who are underprepared for what the world will ask of them.

If AI can write the code, draft the essay, and generate the analysis, what is left to learn? The answer is the hardest thing education has always tried to teach: genuine understanding, deep enough to know when the artifact in front of you is right, and what to do when it isn’t.

In my new Inside Higher Ed piece, I argue that generative AI has broken the old “artifact economy” — the idea that essays, problem sets, and take-home exams reliably signal what a student can actually do.

That problem is obvious. The opportunity isn’t.

The same AI that makes polished artifacts cheap can help us move from grading snapshots to evaluating “game tape” — evidence of reasoning over time, revision, judgment, and growth.

Less: What did you submit? More: How do you think? How do you adapt?

If we get this right, post-AI assessment won’t just survive — it will be more valid and more aligned with what we actually value.

That’s the hopeful fork in front of us.

Read the entire piece here. I would love your reactions.

There’s a moment in almost every hiring and promotion meeting when the room stops talking about the work and starts talking about the person.

It usually happens innocently. Someone flips to the résumé’s summary section—those three confident lines written in the voice of a movie trailer—and reads out loud:

“Strategic thinker. Strong leader. Executive presence.”

Everyone nods, because everyone has learned the choreography. Then someone says a sentence that sounds like analysis but works more like a spell:

“She’s not quite a critical thinker.”

And just like that, a human being becomes an adjective.

I once watched a committee do this with a candidate I’ll call Maya. (Composite story, but if you’ve sat in enough rooms like this, you’ve met Maya.) Maya led a messy cross-functional project: three teams, a slipping timeline, a vendor that kept changing specs, and a VP who wanted “innovation” but also “no risk.” The project shipped. Complaints dropped. Finance—the department that hates everyone—sent a thank-you email.

But in the meeting, nobody opened the notes. Nobody replayed the hard conversations. The group argued about a vibe.

“She’s smart, but does she have leadership presence?”

“I don’t know. She’s… quiet.”

A third person, trying to be kind, added: “She might be a great number two.”

At that point, the meeting is essentially over. Here’s what no one said: Maya ran stand-ups like a metronome. She wrote the decision memo that forced the VP to choose. She escalated the vendor issue at exactly the right time. She ended a conflict between engineering and sales by making them define, in writing, what “done” meant.

Those are not adjectives. Those are actions.

And actions are the only honest language of ability.

Our skill vocabulary is mostly nouns—and nouns are sticky

Most of the way we talk about talent is built from nouns and adjectives: critical thinker, strategic, collaborative, high-potential, data-driven, lacks executive presence. Even when we try to be modern and “skills-based,” we mostly just swap prestige signals for skill labels while keeping the same grammar.

That grammar is imprecise, subjective, and—worst of all—infectious. Once someone is labeled “not strategic,” the label sticks, even though ability is contextual and perishable.

Skills are not medals. They’re muscles. They strengthen with use, shrink with neglect, and look different under different loads. You can be “good at public speaking” one year and terrible the next if you stop doing it. You can become a much better writer in six months with feedback and practice.

Résumé culture didn’t invent this. It industrialized it.

A résumé is a compressed autobiography written for a stranger who has no time. It cannot show the work, so it turns a messy sequence of decisions into polished nouns:

“I ran a weekly experiment cadence and scaled the winner” becomes “innovation.”
“I gave tough feedback without losing the person” becomes “leadership.”
“I wrote the memo that forced alignment” becomes “stakeholder management.”

Now take that logic, pour it into LinkedIn, add endorsements, and you get a public vocabulary that feels objective while staying wonderfully vague.

Portability is useful. Portability without precision is where bias shows up, looking “professional.”

And then skills analytics comes along… and freezes the problem in amber

In the last decade, a new class of tools has promised to make skills “data-driven” by mining job postings and résumés at scale. Lightcast, for example, maintains a skills taxonomy built from job postings, resumes, and online profiles.

This is valuable work. It can reveal what employers say they want. It can show skill drift over time. It can help educators align programs to market language.

But notice the subtle trap: if you build your “skills ontology” by extracting phrases from job ads, you inherit the job ad’s grammar—noun phrases, buzzwords, and all the soft, flattering fog that organizations use when they don’t want to specify what they actually mean.

Job ads don’t say: “Can write a decision memo that drives alignment under time pressure.”
They say: “Strong communication skills.”

So the system learns “communication.” The cycle continues.

A different definition: ability is doing a task well, in context

Let’s replace the noun.

A skill is not “critical thinking.”
A skill is doing critical thinking—when it matters, with stakes, constraints, and tradeoffs.

So how do we talk about ability honestly?

Option A: Define it as observable behaviors (verbs)

Take “critical thinker.” What does it look like when it’s real?

It looks like someone who:

• Surfaces assumptions.

• Tests claims with evidence.

• Separates signal from noise.

• Makes reasoning visible (in writing, models, or experiments).

You can argue about the list. Great. The moment you move from noun to verb, you’re forced to get specific—and specificity is where fairness begins.

Option B: When behaviors vary, define tasks + outcomes

Sometimes you shouldn’t legislate the exact behaviors, because contexts vary and creativity is required. In that case, define the ability by the task and what “good” looks like.

Instead of “strategic,” say:
Task: Choose a 12–18-month product direction under uncertain demand.
Good looks like: Tradeoffs explicit, stakeholders aligned, milestones measurable, plan revised when evidence changes.

Instead of “strong leader,” say:
Task: Run a team through a high-conflict priority change.
Good looks like: People understand why, roles are clear, delivery stabilizes.

Not perfect. But anchored to reality. It’s about the work, not the aura.

The hard part is granularity

In principle, every meaningful activity inside a job is its own ability:

• Can you negotiate scope creep with a vendor without poisoning the relationship?

• Can you debug a production incident at 2 a.m. while keeping stakeholders calm?

• Can you coach a junior colleague through a mistake in a way that makes them better, not smaller?

• Can you write a one-page narrative that a VP will actually read—and then act on?

That kind of specificity is what “ability” really looks like in the wild. It’s also impossible to run as a universal language. Too many edge cases. Too much context. Too much combinatorial explosion.

So we do what organizations always do when something gets messy: we compress. We reach for big nouns: “leadership,” “communication,” “strategic thinking.” They’re portable, which is exactly the problem. The more they travel, the more they mean everything and nothing at once.

What we need is a middle layer: definitions that stay tied to observable behaviors and outcomes, but are expressed at a level that can move across teams, roles, and employers without evaporating.

We can still say “she is a critical thinker,” but only if we can unpack it into observable component behaviors—and then show evidence of those behaviors in action. This is where AI and the “movie of work” ideas I discussed in a previous article become useful: not to replace judgment, but to capture, organize, and surface the behavioral proof that our labels are supposed to represent.

None of that works, though, if the language is sloppy. AI can’t evaluate fog. It needs a vocabulary it can match evidence against—and humans can recognize as fair.

The wiring problem

There’s some good news: the serious work classification systems—O*NET in the U.S., ESCO in Europe, SkillsFuture in Singapore—already lean toward verbs. Many of their skill definitions break down into lists of specific behaviors, not just vague labels.

But they’re missing the connective tissue.

These systems (and, for that matter, most job descriptions) tend to list tasks and skills side by side, like two separate inventories: “Here are the tasks for this job. Here are the skills for this job.” They’re cataloged in parallel, like ingredients on one shelf and finished dishes on another, with no recipe connecting them.

That’s the problem. Skills aren’t a separate thing. They’re what shows up while you’re doing the tasks. You can’t hire for “communication” in the abstract—you need to know: Can this person explain a technical decision to a non-technical executive? Can they do it in writing, in one page? Can they make it clear enough that the executive actually acts on it?

The question a hiring manager actually needs answered isn’t “Does this occupation require communication skills?” It’s more specific: “Which tasks in this role depend on communication? What does good communication look like in this context—when you’re translating technical choices for business stakeholders, with limited time, and real money on the line?”

And now the world is moving from jobs to tasks

This matters more because the labor market is changing its unit of analysis.

More organizations are planning work not as fixed “jobs” but as bundles of tasks and outcomes that can shift among people, teams, contractors, and now AI systems.

You can see the shift in small places first: project marketplaces that staff work in weeks, not years; performance reviews that increasingly cite artifacts (memos, dashboards, recordings) rather than memories; teams that split deliverables between humans and copilots. The unit that matters is no longer “Who are you?” but “What can you do next, and under what conditions?”

Once you see work that way, the missing linkage becomes obvious: if work is modular, your language for ability must be modular too.

You can’t run a task-based world on adjective-based language.

An agenda for getting this right

If we want a labor market that rewards evidence rather than aura, we need a practical language for ability—one that defines what people can do in terms of behaviors and outcomes, not flattering nouns. It has to be simple enough to scale, concrete enough to assess, and flexible enough to travel across roles and industries. Here’s a workable path:

1. Build libraries of task families—the recurring verbs of modern work (diagnose, persuade, prioritize, write, forecast, negotiate, design, coach, debug).

2. For each task family, define a handful of observable “tells”—the small behaviors that reliably show competent execution (the kinds of moves good managers notice, but rarely name).

3. Make context first-class: stakes, ambiguity, audience, constraints. Context isn’t a footnote; it’s the difference between theater and performance.

4. When “how” legitimately varies, define outcomes plus guardrails—what “good” looks like (quality, speed, reliability) and what “good” must never violate (ethics, safety, customer impact).

5. Create explicit task→ability wiring for roles and projects: which tasks matter most, where the bottlenecks are, and what competent performance looks like in the environments people actually face.

6. Treat every definition as a hypothesis. Pilot it, test it against real decisions (hiring, promotion, development), tighten what predicts performance, and retire what doesn’t.

I’d love to see O*NET, ESCO, SkillsFuture, and other skills taxonomy efforts push harder in this direction—less parallel catalogs of “tasks over here, skills over there,” and more of the connective tissue that links tasks to evidence. And if any private actor is positioned to mainstream this, it’s LinkedIn: it already sits on the world’s largest skills vocabulary, but today that vocabulary is still mostly nouns and adjectives. A shift toward verbs, behaviors, and task-linked outcomes would be a genuine upgrade to how talent is understood.

Why this matters

When we define skills as nouns, we end up hiring for the ability to sound like a noun.

When we define skills as verbs, we can hire for the ability to do the work.

That shift creates a cascade: clearer hiring, clearer development, fairer assessment, less political promotion. Candidates know what to practice because the target is visible. Managers know what to look for because the work is legible.

Maya doesn’t need “leadership presence.” Maya needs a world that recognizes the work she already did—and shows her the next behaviors she can learn.

We don’t have a talent crisis.

We have a language crisis.

And the fix is grammatical: stop worshiping nouns. Start measuring verbs.

To my readers: if you’ve seen or use skills frameworks that already come close—ones that connect abilities to observable task-level evidence in a usable way—please leave me a comment. I’d love to hear about them.

In my last piece, I argued that generative AI is turning the old hiring “snapshots” (résumés, portfolios, polished interviews) into something closer to deepfakes. Artifacts that reveal as much about someone’s tools as their talent. And then I suggested the alternative: stop staring at the snapshot. Watch the work. Build “movies:” short, job-relevant behavioral highlight reels that capture process, not just outputs.

If that sounded like a thought experiment, this is the part where I tell you: it isn’t.

The movie of work is already here. We just don’t call it that.

A “movie of work” is not a camera following you around all day. It’s something much more specific: time-ordered snippets of how someone behaves in job-relevant situations, stitched together into evidence. We already run entire professions on this idea. AI is making this concept applicable to many more domains.

Where The Movie Already Exists

Start with pilots, because aviation is one of the few domains where we long ago accepted a blunt truth: the cost of being wrong is too high for vibes.

The FAA literally has a National Simulator Program with detailed rules for qualifying flight simulation training devices. The profession created a structured, repeatable “movie” of performance—normal procedures, abnormal procedures, emergencies—then decided it was credible enough to certify people with it. No résumé. No interview about that one time you handled turbulence well. Just demonstrate the maneuver. Again. Under observation.

Medicine has been drifting the same way. Simulation-based OSCEs (Objective Structured Clinical Examinations) exist precisely because a transcript doesn’t tell you what you need to know about performance under pressure. High-stakes clinical simulations assess everything from emergency response to breaking bad news to a patient’s family. Even the frontier question—VR-based assessment versus in-person—is being actively studied.

Now move to hiring at scale.

Walmart uses something called a “Virtual Job Tryout”—a multimedia simulation that presents prioritization dilemmas, customer scenarios, and judgment calls. Not a personality test. A synthetic slice of work. Deloitte UK runs candidates through an “immersive online assessment” and job simulation before they ever speak to a human interviewer. Unilever famously pushed applicants through online games and scenario-based video steps, cutting their hiring timeline from four months to four weeks while reaching candidates who would never have applied through traditional channels.

Even hospitality, an industry not exactly known for Silicon Valley hype, has played in this space. Marriott launched “My Marriott Hotel,” a Facebook game that simulated running kitchen operations, handling staff, and managing budgets. It was recruiting disguised as entertainment, or entertainment disguised as assessment. Either way, it was a movie.

And if you want the clearest proof that behavioral evidence beats résumé theater, look at sales. In modern sales organizations, the “movie” isn’t hypothetical at all: calls are recorded, searchable, replayable. Gong literally markets the idea of going back to “watch the game tape”—managers review actual customer interactions the way coaches review fourth-quarter plays. The snippets become coaching material, onboarding templates, examples of what “handling an objection” actually looks like in practice.

So the movie of work is already here. It shows up under different names: simulator checks, OSCE stations, job tryouts, day-in-the-life assessments, call game tape. AI is facilitating the creation of such “movies” and making the idea applicable in an ever-wider range of domains.

A Taxonomy of “Movies”

Think of “movie of work” as a family of formats, not one thing. Some already exist. Some are emerging. Some become possible only when AI starts generating situations, capturing process, and compressing it into something human-reviewable.

Here’s how to map the territory.

1) Candidate-Curated Reels

This is the oldest form: the applicant brings evidence they chose.

Teachers show classroom video clips—ten minutes of a lesson where a student asked a hard question, and you can see how the teacher pivoted. Designers present case-study walkthroughs where they narrate the problem, the constraints, the iterations, and the tradeoffs. Some sales candidates submit mock pitches. Architects arrive with renderings, yes, but also sketches that show the thinking before the polish.

The movie here is edited, which is both the point (it’s a highlight reel) and the vulnerability (it’s curated). But even curated evidence beats an unsupported claim.

What AI changes: it gets easier to verify context (metadata, timestamps, provenance) and easier to standardize how clips are tagged. Instead of “I’m good with stakeholders,” you submit three tagged examples: “objection handling,” “explaining tradeoffs to non-technical audiences,” “de-escalating a tense meeting.”

2) Employer-Generated Simulations

These are the workhorses of modern high-volume hiring: simulate job-like scenarios and log decisions, sequence, and judgment.

Walmart-scale job tryouts live here. Deloitte-style immersive assessments live here. So do the coding challenges at many tech companies—not the LeetCode puzzles that measure algorithmic recall, but the “here’s a messy codebase and a bug report, show us how you debug” problems that actually mimic the job.

The key advantage: simulations can measure process. What you notice first. What you prioritize when everything feels urgent. What you do when information is incomplete. Exactly the things résumés are worst at revealing.

The limitation: they’re expensive to build well, and they don’t travel. Pass Deloitte’s simulation, and it means nothing at McKinsey.

3) On-the-Job “Game Tape”

This is the most underappreciated category because it already has cultural legitimacy in certain fields.

Sports has film. Sales has call recording and review. Customer support teams at companies like Zappos and Ritz-Carlton have long recorded interactions—not for surveillance, but for training. (”Here’s what great sounds like. Here’s how Sarah turned around an angry customer. Let’s study it.”)

Certain law firms review oral arguments. Some teaching hospitals review recorded patient encounters. Software teams use recorded user research sessions as shared artifacts.

This category matters because it proves something subtle: once recording becomes normal inside an organization, the real value isn’t surveillance—it’s feedback, shared standards, and the ability to say “Let me show you what I mean” instead of “Let me tell you what I think happened.”

4) Micro-Demonstrations Over Time

This is the bridge to the future you can feel coming.

Instead of one high-stakes assessment, imagine many small ones across time—cheap to run because AI handles setup, adapts difficulty, and provides feedback.

This could look like: a weekly stakeholder-roleplay run for product managers. A rotating set of “competence playlists” that evolve with the role. Periodic scenario refreshers the way pilots do recurrent training—except for customer success reps, analysts, even managers.

The point isn’t perfection. It’s trajectory. A movie can show growth in a way a snapshot never can. You can see someone struggle with prioritization in month one, get coaching, and handle it smoothly by month three. That narrative has value.

5) The Agent-Compiled Behavioral Reel

This is the direction that excites me the most. The piece people some people quietly dismiss as futuristic—until you spell it out.

Imagine you have a personal work AI agent whose job is not to do your work for you, but to capture and compress your work into auditable evidence.

It watches the workflow exhaust you already create: task tickets, design docs, code reviews, meeting agendas, customer emails, decisions and reversals. It asks you, at low-friction moments: “Was this a hard tradeoff? Did you change your mind here? Who disagreed with you? What did you do next?”

And then it turns that into candidate-controlled snippets. Not a transcript of everything. A highlight reel. “Here are three moments where you handled conflict. Two where you debugged ambiguity. One where you used AI as a tool without outsourcing judgment.”

This is not a bodycam. It’s closer to a producer. It creates a reel that is smaller than the raw footage and more interpretable than a résumé.

This is also the moment portability becomes thinkable, because the unit of value is no longer “the employer owns the raw logs.” The unit becomes “the worker owns the compiled evidence.”

What Still Needs to Be Built

The proto-movies exist everywhere, but they don’t add up to a new trust system because they’re mostly one-off, employer-owned, and non-portable.

To go further, we need four missing layers.

Portability Standards. If every company has its own simulation, its own tags, its own scoring, we’re back to the résumé problem—just with better theater. We need role-level “competence playlists” that are legible across organizations the way pilots share a common grammar of maneuvers and checks. “V1 decision drill” means the same thing whether you’re flying for United or Delta.

Provenance and Trust. If AI can fabricate outputs, it can also fabricate “evidence” unless we build provenance: what was captured when, under what conditions, with what constraints. This doesn’t have to be dystopian. It can be simple: verified scenarios, logged constraints, clear disclosure of tool use, auditability.

Worker-Owned Infrastructure. Right now, most of the “movie” data is vendor-owned or employer-owned. Gong owns your call recordings. HireVue owns your simulation performance. A worker-owned reel changes the bargaining relationship: you can leave and still carry your demonstrated competence—without carrying the company’s confidential raw footage.

A Social Norm Shift. The hardest layer is cultural: accepting that for many roles, it’s normal to evaluate people the way we evaluate athletes, clinicians, and pilots—by watching repeated demonstrations in realistic contexts. Not because humans are machines, but because language is cheap now, and we need a sturdier bridge between claim and capability.

A Note on Guardrails

If you’re already feeling the hair on the back of your neck stand up—good. That’s the correct reaction.

A “movie of work” can become a tool for dignity and mobility, or a tool for surveillance and coercion. The difference is policy, consent, and constraints.

We’ve already seen how fast hiring tech can overreach. HireVue, under pressure and scrutiny, discontinued facial-expression analysis in its screening products—an instructive example of boundary-setting. In a future Substack piece, I’ll go deep on what should be prohibited, what should be auditable, what should be worker-controlled, and what should be illegal.

For now, the point is simpler: trust doesn’t emerge from better sensors. Trust emerges from better rules.

The Punchline

A résumé is a postcard from a place you’ve never been, written by someone who may or may not have visited.

A movie of work is not the truth. It’s still edited, framed, partial. But it has something the snapshot will never have: sequence. You can see what happened first, and what happened next. You can see the moment someone got stuck—and what they did about it.

And once you can see sequence, you can do the thing hiring has always pretended to do but rarely can: you can separate polish from competence.

Which means the question is no longer whether the movie of work is possible. It’s whether we build it as a ladder—or as a leash.

I Want to Hear from You

This essay maps what I’ve found so far, but the landscape is incomplete. I’m building a more comprehensive catalog of “movies of work” in the wild—and I need your help.

Have you encountered behavioral assessment formats I’ve missed? Know of industries quietly running on demonstration-based evaluation? Seen creative examples of “game tape” being used for development or hiring? Come across startups or tools offering new ideas in this space?

Reply in the comments or email me at dellarocas@gmail.com. If you point me toward something I haven’t covered, I’ll feature it in a follow-up piece—with credit.

The movie of work is already playing. Let’s make sure we’re watching the right footage.

By the time Stephanie reached the 47th résumé, they all sounded the same.

She was a senior recruiter at a fast-growing software company, the sort of place that once prided itself on “hiring for potential.” Over the years, she’d built a mental catalog of signals: candidates who wrote clearly, who could describe a failure without flinching, who showed their work instead of just name-dropping employers.

This morning, her catalog failed her.

Every résumé read like it had been written by the same person on the same very good day. The bullets were crisp. The achievements glowed. The cover letters threaded the needle between confident and humble, referenced the company’s mission with uncanny specificity, and mirrored the tone of the job description.

It wasn’t until Stephanie reached a line that read, “My passion for your company’s journey to redefine productivity aligns perfectly with your bold vision of the future” that something snapped.

She opened the file in a different viewer. Hidden in the metadata: “Generated with assistance from ChatGPT.”

Of course.

Over the next few weeks, she noticed the same pattern everywhere. Portfolios with pixel-perfect case studies that felt strangely generic. GitHub repos with README files that sounded more impressive than the code they described.

And then came the remote panel interview.

The candidate was polished, maybe a bit too polished. He answered complex questions in long, flowing paragraphs that landed with the soft thud of corporate prose. Halfway through, a colleague messaged her privately: “Is he... reading from something? Look at his eyes.”

Stephanie watched the tiny video window. Every few seconds, the candidate’s gaze flickered just off-screen, as if consulting a silent partner. It didn’t take much imagination to picture a second window open next to Zoom: a chatbot feeding him lines in real time.

For years, recruiters like Stephanie tried to read between the lines of messy résumés and awkward interviews. Now the lines themselves were changing. The proxies they had always used—words, polish, fluency—were being mass-produced.

Her nose for talent no longer worked.

And here’s the uncomfortable part: The system wasn’t suddenly breaking. It was finally being exposed.

The Great Polishing Machine

We like to imagine that AI will give us better ways to see talent. Eventually, it might. But in the short term, it’s doing something simpler and more destabilizing: it’s taking the small advantages a few people had in presenting themselves, and handing them to everyone.

The person who grew up with a parent in consulting and knew how to talk about “impact” and “stakeholders.” The colleague who could turn any task into a heroic story.

Those used to be slightly unfair skills. Now they are buttons and prompts.

Type “rewrite my bullets for a product manager role,” and an AI model obliges. Paste in a job description, and it tailors your résumé, reframes your internships, and manufactures a narrative arc out of a summer job.

From the recruiter’s perspective, the effect is the same: the words on the page no longer belong entirely to the person whose name is at the top.

AI removes the friction from gaming résumés. When everyone has access to the same polishing machine, polish stops being a differentiator. And yet, for lack of anything better, polish is exactly what many hiring processes still reward.

When the old signals are cheapened, what do we fall back on?

The Snapback to Prestige

The simplest answer—the one we’re already drifting toward—is that we fall back on what feels safest: pedigree.

If every résumé sounds good, you start asking different questions: Where did they go to school? Which brands have they worked for? Who referred them?

These were always part of the equation. AI just increases their relative weight.

The danger is that companies respond to the flood of polished signals by doubling down on the least fair filters they already have. Instead of thinking, our proxies are breaking; maybe we need better evidence; they think our proxies are breaking; maybe we need stricter proxies.

So, the degree from a prestigious university matters more, not less. The first job at a recognized brand becomes a golden ticket. Internal referrals carry more weight.

In other words, AI doesn’t just make it easier to fake signals; it nudges the whole system toward credential conservatism. The outcome is predictable: nontraditional candidates get squeezed harder, late bloomers find even fewer on-ramps, and someone’s actual work life is reduced to their first few frames—where they studied, where they started, who opened the first door.

The irony is that generative AI, which we like to talk about as a democratizing force, can easily end up reinforcing the very hierarchies it was supposed to disrupt. Not because of some innate technological bias, but because of how we respond to the uncertainty it creates.

An Economy of Snapshots in a World of Deepfakes

There’s another way to describe what is happening.

For decades, our hiring systems have relied on snapshots: static artifacts that stand in for a person’s ability. A test score, a diploma, a one-page résumé, a 30-minute interview.

AI is quietly turning snapshots into something closer to deepfakes; not maliciously fabricated, but artifacts that tell you as much about someone’s tools as they do about their talents.

Is that blog post entirely their writing, or did they start from an AI draft? Is that code sample entirely their logic, or did Copilot scaffold half of it?

In a world where tools and humans co-create, the old question—” Did they do this alone?”—stops making sense. But our evaluation systems haven’t caught up. They’re still built for snapshots that assume a clean separation between the person and their environment.

The gap between what we think we’re seeing and what we’re actually seeing widens.

The Other Path: Watching the Work

There is, however, another way this could go.

If AI is good at manufacturing polished artifacts, it’s also good at something more constructive: generating situations.

In sports, we trust game tape because it shows behaviors that matter to the game. We watch how an athlete moves, reacts, anticipates. We’re not guessing from adjectives; we’re watching the work.

For most jobs, we’ve never had an equivalent.

Now imagine something different. Instead of sending a résumé to a hundred companies, a candidate logs into a platform offering realistic challenges: a product manager simulation with conflicting stakeholder demands, a customer support scenario with escalating messages, a data analysis task with messy data and ambiguous instructions.

Some simulations would be company-specific. Others would be standardized by third-party platforms or industry associations. And some scenarios would come straight from the candidate’s real work—short vignettes captured and structured by AI from past projects.

AI can spin up and adapt these situations, play the roles of difficult stakeholders, and generate infinite variations. More importantly, it can record the process—not just the final answer, but the path: Which questions did you ask? How did you prioritize? Where did you get stuck? Did you blindly accept the first plausible suggestion, or did you test and refine?

The result isn’t a snapshot. It’s a highlight reel, a “movie” of how you think and act in a situation that resembles real work.

“But Won’t People Just Use AI There Too?”

Of course they will.

The point isn’t to build a world where no one ever leans on AI. That world is gone. The point is to shift what we’re measuring.

Today, we mostly see outputs: the final résumé, the coached interview performance. AI is spectacular at fabricating outputs.

In a well-designed simulation, the signal moves from “Did they use AI?” to “How did they use it?”

Two candidates might both have a chatbot window open. Their traces will look very different: One pastes the entire prompt and copies the answer. The other asks targeted questions, tests suggestions, rejects bad ones, and stitches the good ones into a context-aware solution.

Over multiple scenarios, patterns emerge: Do they notice when the AI is obviously wrong? Do they use it to explore options, or as a crutch? Do they get better at orchestrating human judgment and machine assistance?

In a world where everyone will work with AI, that is not cheating. That is the job.

We stop asking, “Did this person work alone?” and start asking, “Can this person work effectively with powerful tools without outsourcing their brain?”

How AI Can Actually Fix the Trust Problem

If we follow this path, AI doesn’t just break the old trust stack; it helps us build a better one:

From one-shot tests to continuous “micro-tape.” Candidates accumulate many small demonstrations over time—simulations, projects, coding sessions. Each one is cheap to run, because AI handles setup and feedback.

From unverifiable claims to portable evidence. “I led a cross-functional initiative” could link to a bundle of artifacts: decision logs, scenario runs, feedback from stakeholders.

From static labels to dynamic growth. Because simulations are repeatable, people can show improvement. A candidate might include not only their best attempt but also an earlier one, with a visible trajectory. Late bloomers finally get a way to prove that their movie doesn’t end with the first frame.

Over time, industries can converge on common “competence playlists” for roles—scenarios and behaviors that actually matter. The same AI technologies that make résumés untrustworthy can make ability more visible and trust more grounded than it has ever been.

From Snapshots to Movies

This is the fork in the road that AI forces.

One path is easy: we let AI flood the system with better snapshots. We respond by tightening our grip on prestige, network, and gut feel. We call it “raising the bar.” Non-traditional talent disappears further into the background.

The other path is harder: we admit that our proxies were always fragile. We stop pretending that résumés and degrees can carry the weight of everything we want to know. We use AI to generate better evidence—situations, simulations, “highlight reels” of actual work—and we build new ways to store and trust that evidence.

We move, slowly and unevenly, from an economy of “snapshots” to an economy of “work movies.”

AI will, likely, break hiring before it fixes it. Whether that pushes us backward—to more prestige and more guesswork—or forward—to better evidence and better language for what we see—isn’t a technological decision.

It’s a choice about what merit means, and it’s ours to make.

Imagine if we recruited athletes the way we hire employees.

We’d ask them where they went to school.
We’d have them write essays about teamwork.
We’d schedule thirty-minute interviews to discuss hypothetical game situations.
Then we’d pick the ones who seemed like a good fit.

Ridiculous, right? Yet that’s exactly how we scout for talent outside sports, in offices, universities, and boardrooms. We ask for credentials instead of calling plays. We weigh polish over performance.

In sports, guessing is heresy. The stopwatch, the game tape, and the scoreboard don’t lie. Outside sports, guessing is policy.

The Economy of Observable Skill

Every year, thousands of young athletes line up to prove themselves at the NFL Combine. They sprint forty yards while lasers clock every millisecond. They leap, lift, and pivot under the watch of high-speed cameras. Scouts take notes. Analysts build models. The athletes’ résumés — what college they attended, who coached them — fade into background noise.

Because in that moment, the only question that matters is: Can they play?

It’s a brutal meritocracy. A first-round pick can flame out. A walk-on can become a starter. A late-round quarterback from a small Midwestern college can outshine a Heisman winner if he delivers when the ball is snapped.

Contrast that with how most hiring decisions are made. We invite candidates to describe what they’ve built, but evaluation still tilts toward where they built it and how well they narrate it. We lean more on proxies: GPA, degree, company logos, LinkedIn endorsements, all polished symbols of potential that may or may not correspond to actual skill.

We’re scouting blindfolded, relying on intuition instead of observation.

What Sports Gets Right

The beauty of sports is not that it’s perfectly fair — it isn’t — but that it’s transparent. Everyone knows what counts, and everyone sees the same scoreboard.

1. Defined skills and roles.

Every position has a behavior-based blueprint. A point guard must orchestrate; a striker must finish; a goalkeeper must anticipate. Coaches and scouts don’t evaluate vaguely; they evaluate fit to function. That clarity turns performance into a shared language.

In most workplaces, by contrast, the expectations are fuzzy. What does it mean to be a “strategic thinker” or a “good communicator”? Without shared definitions, evaluations drift toward impressions.

2. Continuous assessment.

Athletes aren’t evaluated once; they’re assessed constantly; every practice, every game, every season. Performance data accumulates over time, revealing patterns of growth or decline. That makes improvement visible. Development becomes the story.

Outside sports, feedback is sporadic and usually backward-looking. Annual reviews masquerade as insight when they’re really postmortems. We treat ability as static, a credential you earn early and carry forever, rather than something that unfolds and can be refined.

3. Open pathways.

Sports honors the unexpected. You can rise from obscurity. Players from tiny colleges or overseas leagues can make the big stage, not because someone vouched for them, but because they proved it. Performance can override pedigree.

In most careers, the door closes early. If you didn’t attend the right school or land the right internship, the system rarely lets you demonstrate how far you’ve come since. We’ve built a culture that celebrates potential but distrusts improvement.

4. Transparent evidence.

Every play is recorded. Every stat is public. You don’t need to “trust the résumé”; you can watch the film. That visibility allows accountability.

The rest of us work in opacity. Our output lives behind emails and meetings. Success depends on narratives rather than evidence; stories told in performance reviews rather than performances themselves.

Lessons for the Real World

Not every job lends itself to a stopwatch. We can’t time empathy or grade creativity on a curve. But we can observe behaviors that express those traits and make a difference to the outcome.

The lesson from sports isn’t about statistics; it’s about design. Sports builds systems that directly observe behavior that matters to the game. That visibility changes everything: how coaches develop players, how fans debate performance, how teams invest. It creates a loop between measurement and improvement.

Outside sports, we break the loop. We credential, often based on opaque criteria, and observe later — if ever. We act as if human potential can be inferred from surface signals rather than revealed through context. And then we wonder why we keep missing out on the next undrafted MVP.

Measure What Matters

There’s a story from baseball that captures this perfectly.

In the early 2000s, the Oakland A’s were one of the poorest teams in Major League Baseball. Their general manager, Billy Beane, realized he couldn’t outspend the Yankees — but he could out-measure them. Using data, he identified players who produced results even if they didn’t look like stars. The revolution became known as Moneyball.

What’s often forgotten is that the system Beane disrupted was also relying on measurements, albeit the wrong ones. Scouts relied on how a player looked: his swing, his stride, the sound of the bat. They mistook style for substance.

That’s the point: even inside sports — the world that supposedly “gets” measurement — systems drift toward proxies if the design invites them to. Baseball had to redesign its evaluation infrastructure to measure the behaviors that actually matter.

Sports learned. Most other fields haven’t.

From Snapshot to Movie

The challenge isn’t to measure more; it’s to measure better — to see ability as unfolding context rather than frozen verdict.

A credential is a snapshot: where you went, what you scored, who recommended you.
Performance is a movie: what you did, how you adapted, what you learned.

Sports runs on movies. We can watch the tape, frame by frame, to understand not just what happened but why. Education and employment, by contrast, freeze people in a single frame.

The irony is that modern technology makes continuous observation easier than ever. AI can simulate “game situations” for almost any skill. It can record decisions, highlight patterns, and surface evidence of judgment, the very thing that résumés can’t show.

The future of talent isn’t in better filters. It’s in better film.

The Playbook for a Performance Culture

So, what can the rest of us borrow from the world of sports?

1. Define what good looks like

Every role should have a clear “game tape.” What behaviors separate the good from the great? What decisions matter most? Without definition, measurement is arbitrary.

2. Give people chances to perform

Instead of relying on interviews, design simulations, projects, and trials that mirror real work. Musicians audition behind curtains. Coders show their GitHub. Designers share portfolios. Imagine if educators, consultants, or managers could build similar “tapes” — records of real decisions under pressure.

3. Make evidence portable

In sports, performance travels with you. A player doesn’t start from zero every time they switch teams. Their stats, clips, and records form a living résumé. Imagine if other industries had that: verifiable, portable records of competence. It would make credentials less about prestige and more about proof.

4. Update assessments continuously

Ability isn’t a snapshot; it’s a movie. A test taken at age twenty-two shouldn’t define you at forty-two. We need systems that track progress, not pedigree.

5. Leave room for surprise

In sports, the greatest stories are upsets. The same should be true elsewhere. A healthy system must allow for new entrants, for people whose ability emerges later, or who didn’t follow the scripted path.

Seeing What’s in Front of Us

We keep talking about “finding talent,” as if talent were a hidden treasure. But maybe it’s not hidden. Perhaps it’s just unseen, buried beneath the proxies we trust.

Sports doesn’t ask for belief. It asks for evidence. And evidence, once visible, changes everything: who gets trained, who gets promoted, who gets believed in.

Imagine an economy that worked the same way, where we trusted performance over pedigree, context over credentials, movies over snapshots. Where ability wasn’t something we assumed, but something we could see.

Proof lives in small, decisive behaviors that turn risk into results when the game is on the line. Sports already knows to watch the tape. Maybe it’s time the rest of us did, too.

On a crisp California morning, sunlight glints off the red-tiled roofs of Stanford University. Tourists wander through palm-lined courtyards. Students jog past sandstone arches. Everything—from the manicured lawns to the humming laboratories—radiates success. Parents on campus tours nod approvingly, as if the quality of an education could be measured in acres of landscaping and the shimmer of glass façades.

It can’t, of course. But we act as if it can.

Higher education has become a faith-based market. We pay extraordinary sums for an experience whose value is almost impossible to verify. Economists have a term for this: a credence good, something you must believe in because you cannot easily measure its quality even after you’ve consumed it. Like a mechanic’s advice or an “organic” label on fruit, a college degree requires trust. Did you really learn what you need to succeed? Did the university actually make you better, or simply busier? The truth is, most of us don’t have the answer until many years after we graduate. And because we don’t, we substitute belief for evidence.

In a market built on belief, reputation becomes currency. And in the absence of transparent public measures of learning—how much students grow in knowledge, skills, or judgment—prestige steps in as the universal proxy for quality. It is a proxy that distorts everything it touches.

When Image Becomes Evidence

Universities learned long ago that genuine academic quality is difficult (though not impossible) to display. Learning happens in private—inside the mind—and cannot be photographed. But stone, glass, and stadium lights? Those make wonderful billboards.

Stanford’s sweeping campus, the University of Chicago’s $81 million Mansueto Library, Louisiana State’s $85 million Tiger Stadium expansion, and Boston University’s gleaming $305 million Data Sciences Center are all examples of this logic: if quality can’t be measured, it must at least look impressive. Architectural spectacle becomes a statement of seriousness. A football championship or a solar-powered skyscraper signals excellence more loudly than an incremental improvement in pedagogy ever could.

Yet these icons of prestige rarely improve what or how students learn. They are monuments to perception, a way to make excellence visible when genuine learning remains invisible. Over time, that logic hardens into a trap: once one university builds the glass-domed library, the others must follow, not to teach better but to avoid looking worse.

The Arms Race of Amenities

Administrators like to say that students “demand” luxury dorms, gourmet dining, and Olympic pools. But the causality runs both ways. Institutions compete for the same tuition dollars, so they add amenities to signal distinction; students, confronted with price tags rivaling a mortgage, begin to expect corresponding opulence. Arizona State University’s Sun Devil Fitness Complex—with its lap pools, climbing walls, and mindfulness pods—was built to prove that a public university could deliver the same “experience” as an Ivy. The result: another escalation in the college-as-resort competition.

Every new feature creates administrative gravity: staff to maintain, market, and regulate it. Costs rise. Tuition follows. Students then use those higher costs to justify expecting even more. It’s the luxury feedback loop of higher education: a self-perpetuating cycle of spending as signaling.

The Chivas Regal Strategy

Sometimes, prestige isn’t built with marble but with a price tag.

In the late 1980s, George Washington University’s president, Stephen Joel Trachtenberg, made a counterintuitive discovery. He raised tuition dramatically, simply to make GWU look like an Ivy. The result? Applications soared. Families assumed the higher price meant higher quality. The revenue boom funded new buildings and faculty, which further reinforced the illusion. GWU’s reputation climbed; its costs did, too.

Economists call this the Chivas Regal effect, after the Scotch brand that sold more by raising its price. When quality is hard to assess, people assume expensive equals good. Higher education, the quintessential credence good, proved fertile ground for the illusion. Universities across the country followed suit. Today, even mid-tier colleges use a similar trick: post a stratospheric “sticker price,” then discount it selectively through scholarships. The nominal price sustains the aura of exclusivity while keeping enrollment viable through hidden rebates.

The outcome is a market that is both opaque and absurd. Families shop for colleges without knowing the real price until months after applying, while institutions signal quality through numbers that bear no relation to actual cost or educational value. The theater of prestige continues, one financial illusion at a time.

The Tyranny of the Rankings

When markets grow confusing, intermediaries emerge to simplify them. For higher education, that role fell to U.S. News & World Report, whose annual rankings began in 1983 as a helpful consumer guide and metastasized into the industry’s scoreboard.

Lacking easily accessible measures of learning outcomes, the rankings rely on inputs—spending per student, faculty salaries, alumni giving rates, acceptance rates (Note: some of these inputs (e.g., acceptance rate, alumni giving) have been reduced or removed in recent revisions). In other words, they reward how much money a university burns, not how much learning it creates. To climb the rankings, schools must spend more—on faculty pay, facilities, and marketing—creating the perverse incentive that efficiency is penalized while extravagance is rewarded.

Consider the feedback loop this sets in motion. A college invests heavily to raise its rank. The higher rank attracts more applicants, allowing greater selectivity, which in turn improves the rank again. Donors see success and contribute more. Tuition rises to cover the growing overhead. The university becomes “better” in every measurable way, except the one that matters most: how much students learn.

Rankings flatten complex realities into a single number and, in doing so, shape behavior across an entire sector. They make prestige quantifiable, portable, and addictive. And because everyone plays, no one can afford to quit.

The Cult of Selectivity

If prestige cannot be built by spending alone, it can be purchased through scarcity.

Imagine a college admissions officer whose goal is not only to enroll great students but also to reject as many as possible. Every additional application, solicited through glossy brochures and personalized emails, lowers the acceptance rate and raises the school’s standing. It’s a system where inclusion masquerades as access but serves exclusivity.

Selectivity has become higher education’s most potent status symbol. Lacking credible ways to demonstrate what they teach, elite universities demonstrate who they choose. A low acceptance rate becomes shorthand for excellence; a famous brand becomes a proxy for talent. Employers, faced with a flood of applicants and no standardized competency data, happily outsource their screening to the admissions offices of Harvard, Stanford, or Princeton. Prestige, once a reflection of learning, now precedes it.

This dynamic distorts incentives all the way down the pipeline. Families invest in test prep and résumé-polishing from middle school. Anxiety and inequity soar in tandem. Institutions claim success not by how far they move students but by how little movement their students require. The best colleges, paradoxically, are those that need to teach the least.

The Price of Distrust

All this spending and signaling has not produced trust; it has eroded it. As tuition outpaces inflation year after year, legislators and taxpayers ask the obvious question: What are we paying for? Lacking transparent outcome data, governments respond the only way they can: through oversight.

The result is a thicket of compliance mandates. Universities now track everything from crime statistics and graduation rates to debt-to-income ratios and Title IX investigations. Each new rule demands new staff, new systems, new reports. Accreditation, originally meant to guarantee quality, has become another bureaucratic labyrinth of self-study documents and site visits. The regulatory burden, while intended to ensure accountability, feeds the very administrative bloat it seeks to control.

It’s an expensive substitute for trust. When institutions can’t prove their value, someone else will try to verify it for them, at a steep cost.

The Innovation That Couldn’t Speak

Every so often, someone tries to escape the prestige trap. In 2012, Silicon Valley entrepreneur Ben Nelson launched Minerva University, a radical rethinking of higher education with no campus, no traditional lectures, and a curriculum built entirely around cognitive skills like critical thinking and problem solving. Students move between cities worldwide, learning online in interactive seminars designed around evidence from learning science. By several accounts, it works: students show measurable growth in reasoning and adaptability.

Yet Minerva struggles for legitimacy. It lacks the familiar markers of prestige: ivy-covered walls, football teams, endowments. Without recognized metrics for learning outcomes, it cannot translate its success into the language of credibility the system understands.

Western Governors University faced a similar battle with its competency-based model, where students progress by mastering skills rather than logging seat time. The idea is sound; the accreditation framework wasn’t built to recognize it.

Innovation, in other words, can’t thrive without measurement. A system that defines quality by tradition will always treat innovation as deviance.

Why the Prestige Trap Hurts Everyone

The prestige economy isn’t a harmless vanity fair; it’s an engine of inequality. As universities chase rank and reputation, costs climb, access narrows, and students bear the burden. Middle-income families stretch finances to afford the “best” schools. Less-prestigious institutions, unable to compete in the glamour race, either imitate the spending or fall behind. Meanwhile, the public’s faith in higher education erodes. Gallup polls show declining confidence in universities across the political spectrum. Legislators question their value; donors redirect their gifts; prospective students weigh whether college is “worth it.”

Even for elite institutions, prestige is a treadmill that never stops. To maintain their status, they must spend ever more, court ever richer donors, and admit ever fewer students. Prestige, once gained, is never secure; it demands continuous tribute.

A Way Out

There is, however, a path forward, one that could free everyone from the spiral.

If prestige is a proxy born of ignorance, the antidote is information. Universities could reclaim their integrity by defining, measuring, and publishing the specific competencies they promise to advance. Instead of marketing “excellence,” they could report evidence: how students’ writing improves from freshman to senior year; how their problem-solving, teamwork, and ethical reasoning evolve; how graduates apply their learning in the world.

A transformation like this isn’t without precedent. Half a century ago, the world of investing operated on prestige and mystique. Investors entrusted their money to the star managers of blue-blood firms, men in tailored suits whose reputations were built on charisma and pedigree rather than transparent performance. Then came the data. As researchers began to publish long-term comparisons between actively managed funds and market indexes, the façade cracked. The index fund revolution, led by Vanguard and later amplified by Morningstar’s performance databases, replaced the alchemy of prestige with the clarity of outcomes. The market learned what higher education still resists: when quality becomes measurable, trust no longer has to be blind.

Higher education could follow the same trajectory. This shift would reframe universities from credence goods—something you must take on faith—to experience goods, where outcomes are observable. Imagine if every college provided a transparent “learning impact report” akin to an investment prospectus: here’s what our students gain, here’s how we know. Institutions that deliver genuine progress would no longer need to compete through spectacle. Regulators could scale back intrusive oversight, confident that trustworthy data exist. Employers could hire based on demonstrated competencies rather than brand names. And students could choose colleges for fit and growth rather than for the gloss of prestige.

Why This Time Is Different

I’m not the first to argue that we should do a better job of measuring what students learn. In the recent past, several national and international efforts have tried—and failed—to bring that vision to life. Does anyone still remember the U.S. Spellings Commission or OECD’s AHELO initiative? Their failure wasn’t only technical. It was political, cultural, and deeply institutional. To measure learning meaningfully would require changing not just how we teach but how we define institutional excellence. It would expose uncomfortable truths and upend the pecking order of higher education. And so, politely, the system declined.

Meanwhile, the labor market played along. Employers, needing quick ways to filter candidates, leaned on university names as shorthand for ability. In doing so, they spared academia the pressure to prove what it delivers. Prestige became a convenient fiction that suited both sides: universities sold it, and employers bought it.

But the conditions that sustained this equilibrium are crumbling. The cost of maintaining the prestige economy has become unsustainable—not just for students and families, but for the institutions themselves. Public trust in higher education is eroding, and employers are finding it increasingly difficult to identify real talent by following the familiar trail of credentials and prestige. A quiet but decisive shift is underway.

If universities cannot or will not produce transparent, credible measures of learning, employers will. The next revolution in education may not come from within academia but from the marketplace that depends on its graduates—through competency-based hiring, skill passports, and data-driven validation of human ability. The difference this time is that mistrust has reached critical mass, technology has made new forms of measurement possible, and the old equilibrium—of prestige standing in for proof—can no longer hold.

Breaking the Spell

The prestige trap has persisted not because societies are vain but because universities left them no alternative. In the absence of credible outcome measures, reputation was the only currency available. But currencies can change.

Just as transparent return data stripped the mystique from investment firms, reliable outcome measures could restore trust in education. When investors could finally compare funds by performance, the star manager era faded, costs plummeted, and access widened. The same could happen here. The technologies that built today’s data-driven economy can help education finally measure what matters: the expansion of human capability.

Prestige dazzles; evidence liberates.

When universities compete on measured learning rather than perceived grandeur, everyone wins—students, employers, governments, and the institutions themselves. The marble façades can remain, but for the first time, we’ll know what lies behind them.

In future posts, I will drill down on who is already working on the changes I describe in this post. Among other topics, I will cover:

• Skills-based hiring, up close: what’s working, what isn’t, and why most experiments remain boutique.

• Speaking different languages: the fractured world of competency frameworks—and why building a shared vocabulary may be the hardest task of all.

• From courses to competencies: what competency-based education gets right, and why it hasn’t yet broken through.

• Measuring the unmeasurable?: can creativity, judgment, and ethics be assessed with rigor—and can AI help without amplifying bias?

• Navigating the broken system: how learners and workers can protect themselves while reform lags.

Let me know what else you would like me to write about. And I hope you join me as I dig deeper!

Leave a comment

On a warm spring morning in 1482, a young man from Florence sat down to write a letter. He had heard that Ludovico Sforza, the Duke of Milan, was looking for a military engineer. So, he began listing his abilities: he could design bridges and fortifications, drain moats, and construct cannons. He could make weapons that would terrify the enemy and machines that would astonish the world. At the very end, almost as an afterthought, he mentioned that he could also paint.

That young man was Leonardo da Vinci. His letter still exists, preserved in archives as one of the earliest résumés. It is precise, persuasive, and just a bit theatrical. Leonardo knew exactly what his audience wanted to hear.

Five hundred years later, not much has changed. The modern résumé is still a list, a story of where we’ve been rather than what we can do. It still tries to anticipate what the audience will find impressive. And it still produces the same distortion: an elegant document that conceals as much as it reveals.

A Polite Form of Fiction

The most truthful part of any résumé is the scaffolding: dates, degrees, titles. Everything beneath that, those bullet points of achievement, is a kind of polite fiction.

A weekly meeting becomes “strategic leadership of cross-functional teams.” A simple Excel model is transformed into “quantitative forecasting.” The exaggeration isn’t malicious. It’s ritual. We are expected to inflate because everyone else does.

In 1998, psychologists Frank Schmidt and John Hunter published a meta-analysis that examined eighty-five years of research on how employers select talent. They compared nineteen different methods for predicting job performance. Résumé reviews and unstructured interviews, those mainstays of corporate hiring, ranked near the bottom. The winners were the methods that observed ability directly: work-sample tests, cognitive assessments, structured interviews. The evidence wasn’t ambiguous; it was overwhelming.

And yet résumés endured, not because they worked but because they were familiar.

The Comfort of Convention

Human beings are drawn to order. We like the tidy rectangle of a résumé because it makes sense of chaos. It compresses a messy, nonlinear life into something that fits on one page.

They also offer a shorthand for prestige. A Harvard degree. A McKinsey title. A Google logo. These are the bright flags of status that make one résumé gleam more than another. Employers read them the way collectors read wine labels—half for content, half for reassurance.

The sociologist James Scott once wrote that governments prize what is “legible,” even if legibility sacrifices truth. The same logic governs hiring. The résumé turns talent into something that can be filed, sorted, and scored. It provides an illusion of clarity—and sometimes the illusion feels close enough to reality that no one questions it.

The Illusion of Judgment

Hiring, like so many rituals, flatters our instincts. Managers say they can “spot talent.” They look at a résumé and feel a spark of intuition. They talk to a candidate for fifteen minutes and believe they’ve glimpsed potential.

The research shows otherwise. Unstructured interviews, the free-flowing conversations that dominate corporate recruiting, predict future success only slightly better than random chance. But they endure because they make the interviewer feel perceptive. The act of reading a résumé or conducting an interview simulates judgment. It gives us the satisfaction of choice without the burden of evidence.

There’s also a bureaucratic logic to résumés. They create documentation. They look objective. A hiring manager can say, “We treated everyone the same; we looked at their résumés.” Using a better method, like a skills test, oddly requires more paperwork to prove that it’s valid. The résumé, by contrast, is self-justifying. It’s the “safe” choice, even when it’s wrong.

Technology’s False Revolution

When the internet arrived, it should have liberated us from the résumé. Instead, it trapped us inside it.

When applications went online in the 1990s, recruiters drowned in submissions. So, they turned to the Applicant Tracking System (ATS), software that scans résumés for keywords and sorts candidates into piles.

It was a digital miracle that made the résumé even more powerful. Now, what mattered wasn’t ability but keyword optimization. Candidates began writing for machines.

And the machines learned to reward conformity.

In 2004, economists Marianne Bertrand and Sendhil Mullainathan sent out identical résumés, differing only in the names: some “white-sounding” like Emily and Greg, others “black-sounding” like Lakisha and Jamal. The Emilys got 50 percent more callbacks. Ability was constant; perception did the work.

When self-promotion meets algorithmic filtering, bias doesn’t disappear—it scales.

LinkedIn and the Public Résumé

Then came LinkedIn.

If Applicant Tracking Systems buried résumés inside databases, LinkedIn turned them into theater. Before, a résumé was a private performance; now it’s a public broadcast. Everyone’s career sits on stage, polished under flattering light.

The result was paradoxical. Visibility went up—employers could cross-check titles and dates—but authenticity went down. People curate not just their jobs but their personas: posting thought-leadership essays, collecting endorsements, counting connections. The résumé became a performance art.

Recruiters no longer read; they search. Algorithms rank candidates by keyword density, recency, and reach. The logic of the résumé—presentation over substance—simply moved online and multiplied.

LinkedIn democratized opportunity while standardizing polish. Everyone now plays from the same playbook: the professional headshot, the power verbs, the AI-polished summary. It made us more legible, yes, but not more knowable.

The Final Straw

Enter generative AI.

In the old days, you had to learn how to write a résumé that sounded professional. Today, you paste a job description into ChatGPT and out comes something clean, confident, and blandly perfect.

Recruiters are drowning in these AI-crafted twins: glossy, frictionless, eerily identical. MIT researchers found that candidates who used AI to polish their résumés were eight percent more likely to get offers. One algorithm now writes the résumé; another scans it. Somewhere in between, the human being quietly disappears.

AI was supposed to level the playing field. Instead, it leveled it downward. When everyone looks equally polished, no one stands out.

A Masterpiece Instead of a Résumé

There is, of course, another way.

In sports, recruiters don’t read about performance—they watch it. They time the sprint, record the swing, measure the reaction under pressure. A baseball pitcher doesn’t claim skill; he demonstrates it. Sports scouting, for all its biases, is at least grounded in evidence.

The medieval guilds understood this, too. To become a master carpenter or goldsmith, an apprentice had to produce a “masterpiece”—a tangible proof of skill. Musicians audition. Programmers post code. Designers share portfolios. These are signals tethered to reality, not rhetoric.

Schmidt and Hunter’s research made this point empirically: the closer hiring gets to observing real work, the better it predicts future performance.

So why not apply that principle beyond art and athletics? Imagine if candidates carried competency portfolios—collections of essays, code, designs, analyses, recordings—anything that reveals what they can actually do. Short, targeted assessments could complement these portfolios, offering a glimpse of how people think, solve problems, and adapt under pressure. Together, they would let employers see talent, not just read about it.

But replacing stories with proof is easier said than done. Knowing what to measure and how to measure it has always been the hard part.

The Cost of Change

That difficulty explains why the résumé still rules. Most recruiters, and even many hiring managers, have only a partial view of what predicts success in a role. They’re trained to screen, not to assess. Pressed for time, they reach for what’s quick to compare: degrees, job titles, keywords, years of experience. Even when they know these signals are flawed, they stick with them because the alternatives have always seemed too hard to scale.

But that is beginning to change.

Across industries, new systems are giving organizations a clearer language for describing what competence actually looks like. Competency frameworks such as O*NET, Lightcast’s Skills Taxonomy, and Europe’s ESCO translate job requirements into observable behaviors. Instead of “bachelor’s degree required,” a posting might now ask for something testable—running an A/B test, leading a client workshop, or writing persuasive copy. These frameworks turn success from a credential into a behavior.

Technology is catching up, too. Once limited to pilots and surgeons, AI-managed simulations now let candidates show what they can do in realistic scenarios—a virtual sales call, a code debugging session, a design critique. Platforms like HireVue, Pymetrics (now part of Harver), and Arctic Shores analyze not just what people answer, but how they think: what they notice, how they prioritize, how they adapt. It’s the difference between reading a résumé and watching a mind at work.

Meanwhile, AI-driven talent systems such as Eightfold.ai, Knockri, and Torre.ai are learning to infer skills from the digital traces of people’s real work: projects, certifications, writing samples, recorded interviews. They are still early-stage, but they signal a shift from guessing at potential to observing proof.

Change will be gradual, but for the first time, the infrastructure is forming. The language, the data, and the tools are converging to make hiring for ability—not biography—possible at scale. We are on the edge of a quiet revolution: the moment when evidence finally begins to speak louder than pedigree.

What We Choose to See

We love résumés because they make the complex legible. They give us the illusion that judgment is simple, that human potential can be condensed into neat lines and elegant verbs. We keep them because they flatter our instincts and protect us from blame. And now we’ve trained machines to read them because we cannot imagine hiring without them.

But the résumé has always been a beautiful lie, not because people are deceitful, but because the format itself is. It promises to reveal ability but delivers biography. It measures polish, not potential.

Leonardo da Vinci got the job in Milan, but not because his résumé was persuasive. He got it because he showed the Duke what he could build. The résumé was the prelude. The proof was in the work.

Five hundred years later, we’ve forgotten that distinction. Until we rediscover it, until we learn to value evidence over adjectives, we’ll keep hiring the most convincing storytellers instead of the most capable doers.

A decade ago, higher education was one of America’s most trusted institutions. Now confidence is at historic lows. Gallup reports that barely a third of Americans still say a degree is worth the cost.

The usual story blames universities: too slow, too costly, too out of touch. There is truth in that. But it leaves out half the picture.

Because this is not just a higher education problem. It’s a systemic failure, jointly produced by universities and the labor market, a failure of how we, as a society, define, signal, and recognize human ability.

And today, no one is happy. Employers spend billions on recruiting but complain about bad hires. Graduates carry trillions in debt yet struggle to find work. Universities shoulder the blame for a system they did not break alone.

Trust will return only when all sides change.

When degrees worked

In the middle of the twentieth century, degrees were an elegant solution to an information problem.

Employers couldn’t easily see what a young applicant could do. A degree solved that by acting as shorthand. It was scarce enough to matter, respected enough to reassure, and aligned closely enough with stable career paths that universities could keep curricula in step with job requirements.

Employers hired for potential and trained for everything else. Universities educated broadly, aiming not just to produce workers but to cultivate citizens—curious, ethical, resilient. The degree was a compact: this person has demonstrated the capacity to grow into whatever you need them to be.

It worked—until the world sped up, and the signal began to blur.

When the signal broke

As college became mass education, degrees lost their scarcity. Meanwhile, the labor market transformed. New technologies turned job skills into moving targets—shifting continuously, faster each time. Companies cut their training budgets and expected candidates to arrive fully “job-ready.” That shift raised the stakes of what a credential was supposed to guarantee.

The degree stayed largely the same. It still listed courses, not capabilities. Students could leave the same program with wildly different competencies yet walk away with identical credentials. Transcripts—never a reliable measure of ability—were made even less meaningful by grade inflation.

The signal grew noisier just as employers were becoming desperate for precision. And this is when the labor market rewired the system in ways that deepened the problem.

How employers made things worse

As uncertainty grew, instead of finding new ways to judge ability, most employers leaned harder on the old proxy. And as the number of degree-holders swelled, something subtle but profound happened. Degrees stopped serving as evidence of potential and became filters, blunt tools to winnow the flood of résumés. Prestige went the same way. The names of elite schools, once rare badges of excellence, turned into something flatter: a sorting device.

Soon employers were asking for more degrees for the same jobs—a bachelor’s where a high school diploma once sufficed, a master’s where a bachelor’s used to do. The weaker the signal grew, the harder they leaned on it.

Piece by piece, these choices built a system that prized risk avoidance over talent discovery. Managers stopped asking the hard question, “Does this person possess the abilities to succeed?” and focused on “Will I be blamed if they don’t?” Extra credentials and flashier brand names offered cover. So, they played it safe—even as their “safe” hires quietly fell short.

The deeper flaw: employers are not fluent in the language of skills

Beneath all this runs a deeper, quieter problem: the labor market has never been very good at saying what it wants. Job descriptions are the evidence. They are notorious for being vague, inflated, or contradictory—laundry lists of buzzwords that no single human could embody. They recycle clichés like “self-starter” or “team player” while skipping the hard work of naming the specific abilities that drive success.

Inside firms, the picture is no clearer. Few employers have mapped the skills their best performers possess in ways others could teach or assess. And every company speaks its own dialect: one firm’s “analytical thinking” is another’s “business sense,” another’s “data literacy.” This Babel of skills makes it nearly impossible for universities to align with employer needs—and equally hard for recruiters to assess candidates with precision.

In the absence of clear, shared definitions of competence, degrees became the stand-in. Employers stopped asking if graduates could do the work and started assuming that a diploma at least meant something—even if no one could quite say what.

How academia played along

Universities, for their part, let the ambiguity stand.

They rarely measured student learning in truly meaningful ways, and when they did, they rarely disclosed it. Grades stood in as the only visible marker of achievement, even though they are well known to be a poor proxy for real competence.

Prestige, not outcomes, drove reputation. So, institutions focused on the things that built prestige: research citations, selective admissions, new labs and luxury dorms, climbing walls, and gourmet dining halls. What they paid less attention to was proof that graduates had, and could show, the abilities that mattered. The degree grew more expensive while its meaning grew fuzzier.

And so, the two sides settled into a quiet pact of opacity—employers refusing to clarify what abilities they wanted, universities refusing to clarify what abilities they delivered.

And students—the people the system is supposed to serve—were left holding the risk.

The result: nobody wins

And this is how we arrived at today’s collapse in public trust. We have built an opaque structure where employers can’t find the talent they need, graduates can’t find the jobs they were promised, and universities—while far from blameless—are seen as the face of a failure that runs through the entire system.

Opportunity now flows not to those who are capable, but to those who look credentialed. Wealthier students can stack the résumé with unpaid internships, certificates, and expensive advanced degrees. Everyone else borrows heavily, often only to end up underemployed.

The return to a degree is still positive on average, but it is now highly correlated with family wealth. For the less wealthy, it is often negative.

Inequality widens. And trust erodes even further.

The false dawn of skills-based hiring

If degrees have become blunt, why not hire for abilities directly?

A handful of big firms—Google, IBM, Accenture—have tried. They map roles to competencies, build assessments, retrain recruiters. It’s a promising shift, and it proves the concept—but it also exposes the problem. Those firms can afford the infrastructure. Most employers cannot. Small and mid-sized companies—the bulk of the economy—don’t have the budget to reinvent hiring from scratch.

Until we build a shared language of competencies, widely accepted assessment tools, and portable skill records, skills-based hiring will remain patchy, uneven, and out of reach for most. When deadlines loom, even firms that pledged skills-based hiring fall back to the old filter: the degree, costly and vague, but legible.

The allure—and limits—of alternatives

And what about alternative education? Bootcamps, certificates, and microcredentials promise quick, targeted training—and sometimes even better short-term outcomes—at a fraction of the cost. But they are narrow by design. They work best as complements, adding fresh skills to a broader foundation. What they cannot replace is the immersive experience that shapes the full range of technical, cognitive, and interpersonal abilities today’s worker-citizens need.

And most still don’t offer credible proof of mastery. They function more like course completions than trusted credentials, with little agreement on what “completion” actually means.

The solution, then, isn’t to swap one opaque signal for a thousand fragmented ones. It’s to reinvent the degree so that it finally delivers what the labor market has always assumed it should.

How to fix it—together

Fixing this requires more than reforming universities. It requires reforming employers, too.

Employers must finally learn to name the competencies their jobs truly require and do so in a common language others can use. They must invest in shared, trusted skill assessments—developed in collaboration with educators—rather than building bespoke ones. They must learn to evaluate portfolios, projects, and performance evidence—not just résumés—and rebuild internal training and mobility so managers are rewarded for hiring potential, not just polished pedigrees.

Universities must redesign degrees as maps of competencies, not bundles of courses. They must develop authentic competency assessments—projects, portfolios, simulations—so students graduate with verifiable proof of what they can do. They must align target competencies with real-time labor market data, make internships and apprenticeships core to the experience, and publish graduate competency outcomes transparently so employers can trust what their credentials mean. And for those who remind us that college is meant to shape citizens as well as workers, I couldn’t agree more—but civic and ethical abilities should be treated as competencies too: clearly defined, intentionally cultivated, and credibly assessed.

The hardest work is the work both sides must do together: build a shared language of competencies, co-create credible competency assessments, and develop feedback loops where employers signal what they value and educators shape learning that delivers it. Without that shared infrastructure, skills-based hiring remains a boutique experiment, and degrees remain a vague default. With it, employers will stop guessing, graduates will stop gambling, and universities will earn back society’s trust.

Meaning and mastery

Degrees once gave employers confidence and students opportunity. Now they offer employers false convenience and students risk.

The answer is not to strip education of its meaning. It is to weave meaning and mastery together—to build a system where employers can see what people can do, and people can be seen for who they are.

Only then can the degree become again what it was meant to be:
not a gate,
but a bridge.

What comes next?

This essay is just a starting point. In the weeks ahead, I’ll dig into the forces that broke our talent system—and the experiments trying to replace it. Some of the topics I will cover include:

• Why résumés lie: how today’s hiring practices fixate on surface signals while ignoring the evidence that matters.

• The prestige trap: how the chase for brand and rank inflated costs while draining degrees of meaning.

• Skills-based hiring, up close: what’s working, what isn’t, and why most experiments remain boutique.

• Speaking different languages: the fractured world of competency frameworks—and why building a shared vocabulary may be the hardest task of all.

• From courses to competencies: what competency-based education gets right, and why it hasn’t yet broken through.

• Measuring the unmeasurable?: can creativity, judgment, and ethics be assessed with rigor—and can AI help without amplifying bias?

• Navigating the broken system: how learners and workers can protect themselves while reform lags.

Let me know what else you would like me to write about. And I hope you join me as I dig deeper!

Subscribe now

Leave a comment

Subscribe now

We live in a world of signals.

A diploma hanging on the wall. A polished résumé. A job title at a prestigious firm. These were built to stand in for something deeper: competence, reliability, potential. They were supposed to tell employers who was qualified, who could be trusted, who deserved a chance.

But what happens when those signals no longer match reality?

When the Signal Doesn’t Match the Substance

Consider Aisha. She’s 19, and when a hurricane tore through her community, she coordinated the relief effort, single-handedly and with great success. No degree. No formal title. When she applied for an entry-level project coordinator role at a non-profit, she didn’t even get an interview. Just a few miles away, Edwin, 16, has programmed an AI chatbot that helps immigrants navigate government paperwork. Brilliant, self-taught—yet unseen, because his GPA isn’t high enough for the school’s computer science honors track.

And then there’s Owen: elite schools, polished résumé (with ChatGPT’s help), stellar interview. He gets the job. A few months later, his manager admits, “We should’ve gone with someone else.”

These are not anomalies. They’re symptoms of a broken recognition system. The signals—degrees, résumés, GPAs, titles—that were meant to reveal talent now too often obscure it. They reward polish over substance, pedigree over performance. And the consequences ripple outward.

A Fault Line in Society

Every year, billions are lost on bad hires. Students rack up debt chasing credentials that reveal little about readiness. Employers struggle to fill roles even as graduates sit unemployed. And resentment builds among those shut out of the “degreed elite,” who see the system as rigged against them.

By 2025, credentials have evolved into a powerful social fault line. On one side stand degree-holders, increasingly treated as a privileged class. On the other are the non-degreed, excluded, skeptical, and resentful. The signals we’ve relied on are not just failing—they’re dividing us.

The Credential Crisis

This is the credential crisis: a widening gap between what our signals say and what people can actually do.

• Employers are flooded with applicants who look great on paper but fail in practice.

• Workers with the right skills are invisible because they lack the “right” pedigree.

• Students gamble on costly degrees that don’t guarantee opportunity.

• Society misallocates talent, wasting potential and fueling mistrust.

Opportunity today rests on costly, opaque credentials that shut out skilled people and often mislead those eager to learn. They reveal little about competencies that matter and evolve too slowly for a labor market that changes by the day.

A Different Way Forward

In sports, talent scouts don't care where someone went to college. They watch the swing, the pass, the sprint — behaviors that are directly relevant to success. Why not do the same everywhere else?

The tools that enable that shift are beginning to emerge. AI can evaluate complex thinking in action. Digital platforms can track real performance over time. Simulations can test judgment under pressure. Blockchain can make skill records secure and portable. For the first time, we can build systems where competence in a wide range of domains can be made visible, trustworthy, and owned by the individual.

In such a world, Aisha’s leadership in disaster relief could be verified through team logs, endorsements, and simulations. Edwin’s programming skills could be visible in a portfolio of GitHub projects he’s built. Employers could define roles around competency needs. Educators could design programs around measurable outcomes that match these needs. And individuals could take ownership of their own ability portfolios.

This isn't just a change in how we educate or hire. It's a deeper shift in how we understand human potential. The question is whether we have the courage to abandon the system that's tearing us apart. Because this isn't just about building new tools—it's about shifting mindsets, incentives, and power structures.

What to Expect From This Newsletter

This Substack, The Credential Crisis, is where I’ll explore how we get there. Here’s what you can expect:

• Bi-weekly essays: Sharp takes on why traditional signals like degrees are failing, and what better ones might look like.

• Case studies: Profiles of organizations experimenting with competency-based hiring and learning.

• Interviews & podcasts: Conversations with innovators building the infrastructure for a competency-driven future—from policymakers to technologists, from educators to CEOs.

• Behind-the-scenes insights: Reflections from my forthcoming book Unseen, which makes the case for reinventing recognition around observable competencies.

• Reader conversations: Opportunities to share your own stories of being unseen, misjudged, or overlooked—and how you’ve found ways to prove your ability.

My goal is to build a community of people who sense what’s broken and want to help fix it. Business leaders, educators, policymakers, technologists, students, parents—this crisis affects all of us.

An Invitation

If you’ve ever felt unseen because your credentials didn’t tell your story—or frustrated because somebody else’s credentials told the wrong story—this space is for you.

Because when talent goes unseen, we waste the future we all share.