Hiring Is Broken — Not Because of AI

For a long time, hiring relied on a quiet but reasonable assumption: if someone has seen this kind of problem before and can execute smoothly, they'll probably do fine again.

That assumption gave us resumes. Experience requirements. Degrees. Skill tests. Interviews.

It wasn't perfect. But it worked well enough — because the assumption underneath it was true. If someone could produce quality output, they probably understood the work. The production process required the understanding. You couldn't fake your way through a complex analysis without actually knowing how to analyze. You couldn't write a compelling strategy without actually thinking strategically.

AI broke that assumption. Quietly, completely, and in ways most hiring systems haven't caught up with.

AI didn't destroy skills. Skills still matter. A software engineer still needs to understand systems. A financial analyst still needs to read balance sheets. The work still requires knowledge, judgment, and technique.

What AI destroyed is what skills signal.

And that distinction — between having a skill and signaling a skill — is now the most expensive blind spot in talent acquisition.

The Pattern Everyone Recognizes

Almost every hiring manager I talk to has experienced this in the past year.

The candidate was sharp. The interview went well. The case study was clean. The portfolio looked solid. References checked out. The team felt confident. The offer went out.

Six months later, cracks appear. The same person who seemed so capable struggles with novel problems. They need heavy assistance. They can't explain their reasoning when pushed. When conditions shift, they freeze. By month twelve, the plateau is undeniable. Performance flatlines. The team compensates. The opportunity cost becomes visible.

The post-mortem always sounds the same: "On paper, they looked perfect."

The conclusion is usually vague: "They're not as strong as we thought."

But that's not the real issue. The real issue is that we evaluated the wrong capabilities. We selected for signals that no longer map to durable performance. The instruments were precise. They measured the wrong thing.

What AI Actually Did to Hiring Signals

AI changed the economics of skill expression in four ways simultaneously. Each one individually would be manageable. Together, they collapse the entire signal system.

Knowledge access collapsed. Information that once required years of study or expensive credentials is available instantly. Anyone can sound informed about any domain within minutes. The time cost that once filtered for genuine expertise has disappeared.

Execution cost dropped. Tasks that required deep expertise to produce — code, analysis, writing, design — can now be scaffolded by AI tools. The gap between "can do it with help" and "can do it independently" has widened while becoming invisible from the outside.

Outputs became polished by default. AI-assisted work looks professional. Grammar is clean. Code compiles. Presentations are structured. The surface quality that once indicated craftsmanship now indicates nothing. A first-year analyst and a ten-year veteran can produce the same looking deliverable.

Cognitive effort became invisible. You cannot tell from the output whether someone struggled for hours to understand a problem or copied an AI response in seconds. The fingerprints of thinking have been erased.

What once required internal models can now be externally scaffolded. And here's the critical implication: skill presence no longer reliably implies skill ownership.

A candidate can demonstrate a skill without possessing it durably. They can perform competently in an interview while lacking the internal architecture to perform competently when conditions change.

So when we say we're "testing skills" today, what are we actually testing? Let's be honest about it.

Resume keywords claim to measure competence. Post-AI, they measure surface familiarity — the ability to include the right terminology, which can now be acquired in an afternoon.

Certifications claim to measure validated knowledge. Post-AI, they measure willingness to complete credentialing processes. Whether that correlates with retained understanding is anyone's guess.

Years of experience claim to measure readiness. Post-AI, they measure time elapsed. Ten years of experience and ten repetitions of year one look identical on a resume.

Interview fluency claims to measure intelligence. Post-AI, it measures confidence under ideal conditions — the ability to perform well when questions are expected and preparation is possible.

Case studies and take-homes claim to measure problem-solving. Post-AI, they measure AI-assisted execution — the ability to produce a polished deliverable with unlimited tool access and no time pressure.

Portfolios claim to measure track record. Post-AI, they measure curation skill — the ability to present work attractively, regardless of how much was personally produced.

"Tell me about a failure" claims to measure self-awareness. Post-AI, it measures narrative skill — the ability to construct a compelling growth story, which can be rehearsed and polished with AI assistance before the interview even starts.

References claim to measure real-world performance. They measure relationship management. Candidates choose their references. References have social incentives to be positive.

Notice the pattern.

Every single one of these signals over-weights what candidates know and how smoothly they execute. None of them observe what happens when the candidate is wrong. None of them require revision. None of them measure improvement from attempt one to attempt two.

They capture a snapshot of polished performance and extrapolate from it a prediction of durable capability.

That extrapolation no longer holds.

The Three Capabilities Underneath All Work

Forget job titles for a moment. Forget seniority levels and competency matrices.

Every person performing work — any work, any role — operates through three capabilities simultaneously. They're always present. They're not stages. They're not sequential. They're intertwined in everything.

Recognition

The ability to orient correctly. What kind of situation is this? What matters here? What doesn't? Recognition is pattern-matching against experience. It determines whether you even frame the problem correctly before attempting to solve it.

There's a spectrum: Unaware → Aware → Acquainted → Comfortable → Immersed.

Hiring has traditionally relied heavily on recognition signals — resumes, years of experience, brand-name employers, credentials. The problem is that AI massively compresses this spectrum. People can sound immersed without having built deep internal models. They can recognize patterns without understanding them.

Recognition still matters. It's table stakes. But it no longer differentiates.

Execution

The ability to act smoothly under stable conditions. Given that you understand the situation, can you produce the required output? Execution is fluency — the capacity to move from intention to result without excessive friction.

There's a spectrum: Clumsy → Functional → Smooth → Efficient → Automatic.

Interviews, case studies, take-home assignments — they all try to measure execution. And AI is exceptionally good at inflating it. Plans look clearer. Writing looks sharper. Code runs. Analysis sounds confident. Execution today often reflects tool leverage, not internal capability. Which is why it feels impressive in the interview and fails under pressure on the job.

Adaptation

The ability to detect mismatch, revise approach, and improve across attempts. This is simple to define: what happens when the first attempt doesn't work?

There's a spectrum: Brittle → Reactive → Responsive → Adaptive → Self-Correcting.

Some people defend the first answer. Freeze when contradicted. Wait to be told what to do. Repeat the same mistake after being corrected.

Others notice mismatch early. Revise their approach without being asked. Test alternatives. Don't make the same error twice. The second attempt is structurally better than the first — not just polished differently, but genuinely improved.

This difference predicts learning velocity far better than experience, credentials, or skill lists. And yet hiring almost never observes it.

Why Adaptation Is Now the Only Signal That Survives

The logic is straightforward once you see the economic shifts clearly.

AI commoditizes Recognition. If anyone can acquire surface familiarity quickly, Recognition becomes necessary but not sufficient. It's the entry ticket, not the differentiator.

AI inflates Execution. If anyone can produce polished output with AI assistance, Execution signals become unreliable. High apparent fluency may mask low actual understanding.

AI amplifies contradiction. This one is less obvious but more important. AI tools produce plausible wrong answers. They hallucinate. They miss context. They generate outputs that look correct but fail under scrutiny. Working with AI means constantly encountering mismatch between what the tool produced and what the situation actually requires.

So the differentiator becomes: what happens after the first attempt breaks?

Adaptation is visible in four things:

• Mismatch detection speed. How quickly does the person notice that something is wrong? Do they catch errors before they propagate, or do they present AI hallucinations with confidence?
• Revision quality. When they update, is it a surface fix or a structural change? Do they address the symptom or the cause? Is the revision a re-prompt with slightly different wording, or a genuine rethinking of the approach?
• Error non-repetition. Does the same mistake recur? Or does their behavior actually change based on feedback? This is learning made visible. You can watch it happen in real time.
• Improvement from attempt one to attempt two. Is the second output better than the first? Not just different — better. By how much? This is learning velocity, and it's the single most predictive signal for long-term performance in an AI-augmented environment.

The person who thrives with AI is not the one who gets it right the first time. It's the one who corrects fastest when the first attempt fails.

And in a world where AI assistance makes first attempts look increasingly similar across candidates, the correction — the adaptation — is what separates those who scale from those who plateau.

What's Under the Hood: The Four Moves

Adaptation isn't vague. It isn't personality. It isn't "growth mindset" in the motivational poster sense.

Adaptation is produced by specific, observable cognitive operations. There are four of them. We call them the Four Moves.

Generating Alternatives

When facing ambiguity or failure, does the person generate alternative explanations? Or do they lock onto the first interpretation and defend it? High capability here means considering multiple possibilities before converging. Low capability means accepting the first answer — including the first AI output — without questioning whether alternatives exist.

Revising Beliefs

When evidence contradicts their view, does the person actually change their mental model? Or do they adjust their words while preserving their underlying assumptions? This is the difference between genuinely restructuring understanding and surface compliance — agreeing they were wrong without actually updating how they think.

Connecting Patterns

Can the person apply learning from one context to another? Do they recognize that a supply chain problem and a team communication breakdown share the same underlying structure? Or does every problem feel completely novel, requiring a fresh AI prompt from scratch?

Tracing Consequences

Before acting, does the person trace the downstream implications? Second-order effects? Third-order? Or do they execute immediately on first-order reasoning and get surprised by consequences that were predictable?

These four operations are the engine underneath adaptation. When all four are active, people correct fast, learn fast, and compound their capability with AI. When they're dormant, people produce polished first attempts and crumble when those attempts fail.

For the full deep-dive on how these operators work, how they develop, and why AI threatens each one specifically — read Soft Skills Are Dead. Here's What Actually Matters.

For hiring, what matters is this: the operators are observable. You can watch them activate — or not — in a twenty-minute conversation. And they predict performance in AI-augmented environments far better than any traditional hiring signal.

The Smallest Shift That Fixes Hiring

Here's the thing that frustrates me about this whole problem. The fix isn't complicated. You don't need a new hiring framework. You don't need to rip out your ATS. You don't need to retrain your interview panels.

You need a different moment of observation.

Right now, hiring makes its decision after one attempt. One interview answer. One case study. One take-home. That selects for confidence, fluency, and AI-assisted plausibility. It completely misses adaptation.

The fix is three steps:

1. Observe the first attempt. This is what hiring already does. Let the candidate respond to the problem, present their analysis, walk through their approach. Score it. Fine. This captures Recognition and Execution. It tells you whether they can orient correctly and produce smoothly under stable conditions.
2. Introduce contradiction. This is what hiring almost never does. Change something. Add a constraint that invalidates part of their solution. Present information that contradicts their assumption. Surface an edge case their approach didn't handle. Or — and this is particularly revealing — introduce a deliberately incorrect AI suggestion and see if they accept it or challenge it.
3. Observe what happens next. This is where adaptation becomes visible. Do they detect the mismatch? How quickly? Do they revise their approach, or just their answer? Is the second attempt genuinely better, or just differently wrong? Do they update their reasoning, or just their conclusion? Which of the four moves activate?

That's it.

Same interview. Same case study. Same process. One additional moment of observation.

The difference between "good at interviews" and "good at learning" becomes visible. The brittle performers — the ones who look strong on first attempt but can't adapt — are exposed. The adaptive performers — the ones who might not nail it immediately but improve rapidly — are identified.

And in an AI-augmented world, the adaptive performers are the ones who compound their value. The brittle performers are the ones who plateau.

What We Built

We took this methodology and operationalized it.

The assessment is conversational. About twenty minutes. Works on any device. It feels like solving a problem with an intelligent partner, not like taking a test. It creates conditions where Recognition, Execution, and Adaptation naturally activate and become observable.

Scores are based entirely on behavioral evidence. No self-report. No personality inferences. What you see is what the person actually did under conditions of ambiguity and mismatch.

What you get:

• Recognition signal — how correctly they orient to the situation. This establishes baseline. It must be sufficient, but it's not differentiating.
• Execution signal — how smoothly they produce under stable conditions. This is now understood to be inflatable by AI. It appears strong with assistance but may be fragile without it.
• Adaptation signal — what happens when reality contradicts their first attempt. This is the differentiator. It predicts who scales versus who plateaus.
• Operator breakdown — which of the four cognitive operations are driving or limiting their adaptation quality. This tells you not just whether they adapt, but how they adapt and where their specific growth edge is.

Use this for final-round assessment where mis-hire cost is substantial. For internal promotion decisions where current performance may not predict next-level capability. For high-volume screening with an adaptation filter before investing interview time. For development diagnostics to identify who needs support versus who is ready.

The Math

Here's the part that makes the decision easy.

The fully loaded cost of a mis-hire — direct costs, opportunity costs, team disruption, strategic delay — is conservatively north of $400,000 for a mid-senior role. Some estimates put it at 3-5x annual salary when you include the compounding effect of a wrong person in the wrong seat for twelve months.

The cost of measuring adaptation is twenty minutes.

You're already spending weeks on sourcing, screening, interviewing, reference-checking, deliberating. You're already investing heavily in getting this decision right. And then you're making the final call based on signals that AI has rendered meaningless — while ignoring the one signal that actually predicts performance.

AI didn't break hiring by replacing humans. It broke hiring by making Recognition and Execution look strong even when Adaptation is weak.

Until you explicitly observe adaptation in action — not stories about adaptation, not self-reported growth mindset, but actual behavior under actual contradiction — you'll keep making confident, expensive mistakes.

The organizations that figure this out first will build teams that compound with AI. The organizations that continue measuring the wrong signals will build teams that depend on it.

The difference shows up slowly. And then all at once.

GET IN TOUCH →

Short Version:

• Hiring rests on one assumption: past smooth execution predicts future performance. AI broke that assumption.
• AI destroyed what skills signal, not skills themselves. Skill presence no longer implies durable skill ownership.
• Recognition and Execution are now inflatable by AI. Adaptation is the only signal that survives.
• The fix is one extra moment: observe the first attempt, introduce contradiction, watch what happens next.