The Problem Nobody Wants to Name

Here’s what we don’t talk about: the entire system for identifying human capability is broken. Not slightly off. Not in need of tweaking. Fundamentally, structurally broken.

For centuries, we’ve been trying to answer a simple question: who can do what? And for centuries, we’ve been getting worse at it. Every layer of abstraction we added—degrees, tests, certifications, LinkedIn profiles—moved us further from the thing we actually care about: real capability demonstrated in real situations.

The cost isn’t abstract. Conservative estimates put the global waste from talent misallocation at $800 billion to $1.2 trillion annually. That’s roughly the GDP of Indonesia disappearing every year because we can’t figure out who’s actually good at what.

But here’s what makes this moment different: the old system isn’t just inefficient anymore. It’s collapsing. And we have maybe 3-5 years to build something better before the gap between those with good talent signals and those without becomes permanent.

How We Got Here: Four Ages of Measuring People

Age 1: Direct Observation (Pre-Agriculture)

In hunter-gatherer bands of 20-150 people, talent signaling was simple. You claimed to be a good tracker? Everyone watched you track. You failed? You went hungry. So did everyone else.

This created three critical characteristics that every subsequent system has lost:

Perfect Information Symmetry. Everyone saw the same behavioral evidence simultaneously. No intermediaries, no translation, no filtering.

Immediate Feedback. Skill claims were validated or falsified within hours or days. Not years later when someone finally noticed you couldn’t do the job.

Unified Consequences. Those making decisions about capability bore the direct results. If the tracker was wrong about being able to track, the whole band suffered. No one could issue a credential and walk away.

The system couldn’t scale. But it had one virtue everything since has lacked: you couldn’t fake it.

Age 2: Reputation Networks (Agriculture to 1800)

Communities grew past the Dunbar number. You couldn’t watch every blacksmith work. Solution: reputation passed through trusted intermediaries. Apprenticeship systems emerged—masters validated skills through years of observation, compressed into a journeyman’s certificate.

Signal became separate from behavior for the first time. But it stayed anchored through three mechanisms: small networks where guild masters knew each other and reputation mattered, local consequences where bad craftsmen harmed their own community, and multi-year observation periods where apprenticeships lasted 3-7 years.

First cracks appeared. Credentials could be bought. Nepotism crept in. Guild monopolies restricted access. But the connection between signal and capability hadn’t fully severed yet.

Age 3: The Credentialing Machine (1800-1990)

Industrial Revolution. Factories needed thousands of workers. Urban companies needed clerks from other cities, other countries. How do you evaluate someone you’ve never met?

Answer: standardize everything. Schools, colleges, and testing agencies became credentialing monopolies. Universal signals that could travel.

This solved scale but created three pathologies that still plague us:

The Abstraction Problem. Credentials measured inputs (time in school, courses completed) rather than outputs (actual capabilities, growth patterns).

The Privilege Laundering Problem. Since credentials required access—money, connections, geography—they began measuring advantage more than ability.

The Consequence Separation Problem. For the first time, those issuing signals bore no responsibility for accuracy. If a graduate failed, the employer suffered. The university kept its tuition.

Age 4: Digital Discovery Without Validation (1990-2024)

The internet promised to fix everything. LinkedIn connected 900+ million professionals. Finding candidates became trivial.

Selection got worse.

We solved discovery without solving validation. Faced with 250 applications per opening, employers doubled down on credentials as quick filters—not because they believed in them, but because they had nothing else.

Meanwhile, signal gaming reached industrial scale. Résumé inflation—every job became “strategic,” every project “transformational.” Credential inflation—bachelor’s became necessary for jobs that previously needed high school. Network gaming—LinkedIn endorsements with zero validation. Interview coaching—entire industries teaching people to perform competence rather than demonstrate it.

We built a perfect search engine for a library full of books with incorrect titles.

The Structural Problem: Consequence Separation

Here’s the pattern across all four ages: as signal and consequence separate, signal quality degrades.

Look at what happened:

The pattern is clear. When the person creating the signal, the person using it, and the person who suffers when it’s wrong are all the same—you get accurate signals. Separate them, and quality degrades. Separate them completely, and you get what we have now.

This isn’t a bug you fix with better algorithms or reformed universities. It’s structural. You cannot create accurate signals when those who generate them bear none of the consequences of being wrong.

What This Actually Costs

Let me make this concrete.

The U.S. alone sees roughly 60 million hires per year. Average cost per hire is around $4,700 according to SHRM—but that’s just direct costs. True cost including lost productivity runs 3-4x salary. For a $60,000 role, you’re looking at $180,000+ in total hiring costs.

Now here’s the problem: studies suggest 46% of new hires fail within 18 months. That number is contested, but even if it’s half that, the waste is staggering. A bad hire costs roughly $240,000 when you factor in turnover, disruption, training wasted, and opportunity cost.

Then there’s the invisible cost: capable people who never get discovered because they lack the right signals. Research by Amanda Pallais at Harvard showed that simply providing work history signals to previously unemployed workers dramatically improved their outcomes. The capability was there. The signal wasn’t.

Add it up globally—direct hiring inefficiency, bad hire costs, opportunity costs from misallocation—and you get somewhere between $800 billion and $1.2 trillion annually. That’s not a number pulled from thin air. It’s what happens when you run a global economy on broken measurement infrastructure.

Why This Moment Is Different

The talent signal problem has existed for decades. Three things are converging now that make it both urgent and solvable.

The Credential Collapse Is Already Happening

Between 2023 and 2025, over half of major U.S. employers dropped bachelor’s degree requirements for significant portions of their workforce. Google, IBM, Tesla, Bank of America, Walmart. Multiple state governments.

This wasn’t ideological. It was desperation. Talent scarce, unemployment low, companies couldn’t afford to filter out capable candidates based on arbitrary credentials.

But here’s the crisis: removing requirements doesn’t replace them. Harvard Business School research shows that even when degree requirements are dropped, candidates without degrees still aren’t getting hired at equal rates. Managers revert to informal proxies—prestige company names, personal networks, “culture fit”—that are often more biased than credentials were.

We’ve created a validation vacuum. The old system is dying. Nothing has replaced it.

AI Is Making Credentials Obsolete

Every capability that credentials were designed to measure is being automated. Information retrieval, basic analysis, code generation, writing, data visualization, pattern recognition, research.

What remains uniquely human: adaptive learning, judgment under ambiguity, ethical reasoning, emotional navigation, creative synthesis. Growth velocity—how fast you can develop new capabilities when the old ones stop mattering.

A degree proves you can learn, analyze, and communicate. Those are precisely the skills AI is commoditizing. It says nothing about your adaptive learning velocity, your judgment under pressure, your ability to build trust with people who don’t agree with you.

The credentials we spent centuries building measure exactly the capabilities becoming worthless.

New Measurement Is Finally Possible

For decades, measuring “soft skills” or growth trajectories was economically insane. You needed human psychologists at $200-500/hour, multi-day observation, subjective interpretation. Only elite firms could afford it, and even then only for executives.

That changed in the last two years. Large language models can now extract behavioral patterns from natural conversation—hypothesis generation, update velocity, emotional regulation signals—with reliability approaching human psychologists. What used to cost $500-5,000 per candidate now costs $0.10-1.00.

That’s not marginal improvement. That’s a 1,000-5,000x cost reduction. At these prices, continuous behavioral tracking becomes viable. Weekly check-ins. Real-time analysis of work conversations. Passive capture from existing meetings.

The barrier shifted from “can we measure this?” to “should we measure this?”

What Replaces Credentials

When LinkedIn killed its Skills Assessments in 2023, it revealed something important: no single platform can be both marketplace and validator. LinkedIn’s business model requires engagement. Every fake endorsement creates activity. Truth and engagement are structurally opposed.

The solution isn’t one universal system. It’s multiple competing indices with interoperability standards. Exactly how credit reporting evolved.

The Credit Reporting Parallel

In 1950, credit was chaos. Local assessments, no data sharing, massive inefficiency. Credit mostly limited to the wealthy with existing relationships.

Then FICO and competitors emerged. Standardized methodology. Data aggregation. Interoperability. Competition. Multiple scoring systems competing on methodology while sharing underlying data infrastructure.

It worked because: no monopoly (multiple bureaus prevent capture), standards (common data formats), auditing (regulated and transparent), user control (access and dispute your data), market selection (lenders choose which scores to trust).

Multiple Talent Indices

Apply the same logic to capability measurement:

Academic Growth Index (from universities): Learning velocity, knowledge integration, intellectual humility. Data from classroom interactions, project work, peer collaboration over multiple years.

Performance Excellence Index (from employers): Execution consistency, adaptability, leadership emergence. Data from real work with real stakes.

Skills Growth Index (from assessment platforms): Human dimension capabilities, growth trajectories. Data from structured conversations and behavioral analysis.

Domain Mastery Index (from professional associations): Technical depth, ethical practices, community contribution.

Each measures different things, at different timescales, with different data sources. No single index captures everything. That’s the point.

Why Multiplicity Beats Monopoly

You can’t game all indices simultaneously. Different indices matter for different roles. Competition drives better methodology. If one index gets corrupted, others remain valid. Multiple pathways prevent single gatekeepers from excluding groups.

This isn’t utopian. It’s how every mature measurement infrastructure works. We just haven’t built it for human capability yet.

The Obvious Concerns

“This is surveillance.”

Fair concern. Continuous behavioral tracking could become oppressive. But there’s a difference between monitoring for control and observation for validation.

The safeguards matter: individual data ownership (you own your data, full stop), algorithmic transparency (open methodologies, explainable decisions), revocable consent (withdraw access anytime), portability (export and move between platforms).

Credit scores track behavior too. You control who sees them, you can dispute errors, multiple bureaus compete, methodologies are regulated. Not perfect, but demonstrably better than the alternative of no credit access at all.

“People will game it.”

Of course they’ll try. But gaming requires maintaining fake behaviors over months or years—cognitively unsustainable. Multiple uncorrelated signals mean you can’t optimize for everything simultaneously. And unlike credentials, indices can cross-reference with actual outcomes. If your index says “high adaptability” but you fail every time circumstances change, the index updates.

People know paying bills on time improves credit scores. Is that “gaming”? If you consistently pay bills to improve your score, you ARE creditworthy. Same logic: if you consistently demonstrate growth behaviors to improve your index, you ARE high-growth.

“AI reproduces existing biases.”

It can. But indices can also be MORE fair than credentials. They measure behaviors, not demographics. They focus on trajectory (improvement from any starting point) rather than position (where privilege started you). They can be continuously audited for demographic disparities in ways credentials never were.

Compare to the current system: degrees heavily biased by wealth, networks favor existing elites, “culture fit” maintains homogeneity, résumé gaps punish caregiving. Indices aren’t perfect, but they can be demonstrably less biased—and unlike credentials, they can be improved based on evidence.

What Happens Next

I’m not going to pretend I know exactly how this plays out. But I can see the trajectory.

Over the next 2-3 years, a handful of index platforms will launch and start proving predictive validity. Some employers will pilot them, initially for hard-to-fill roles where traditional credentials have obviously failed. The ones that work will spread through case studies and word of mouth.

By 2027-2028, we’ll probably hit some kind of tipping point. Maybe 20-30% of employers using behavioral indices. Credential requirements will keep dropping. Universities will start partnering with index providers or building their own—the smart ones anyway.

By 2030, indices could be infrastructure. Like credit scores. The idea that we ever hired people based primarily on where they went to school will seem as primitive as bloodletting.

Or we drift into proprietary platforms with opaque scoring and new forms of gatekeeping. Reproduce current inequalities in digital form. Lock in advantages for those who got there first.

Which path we take depends on who builds this and what values they embed. Whether we design for transparency or opacity. Whether we create multiple pathways or new monopolies. Whether we measure growth from any starting point or just reward whoever started ahead.

Subscribe now

Share