I work with AI every day. I build systems with it, argue with it, occasionally watch it produce something brilliant, and frequently watch it produce something that sounds brilliant but is confidently, spectacularly wrong. (If you’ve ever read an AI-generated report and thought “this sounds amazing” and then checked a single number — you know the feeling.)
And the more I work with it, the more convinced I become that if you have 20 or 30 years of real domain experience, AI just made you more valuable.
The Middle Is Collapsing
There’s a layer in every organization that I call the execution middle. You know these people. You might be managing some of them. They’re the ones who take a decision from someone senior and translate it into work — the coordinators, the analysts, the people who turn strategy into tasks, manage queues, route requests, and generally keep the machinery from catching fire.
AI is eating that layer alive.
A Korn Ferry survey found that 41% of employees say their companies have already cut managerial layers. Managers now supervise six people instead of three. The extra responsibility still needs time and attention, even when a layer of management disappears. Gartner predicts that by 2026, 20% of organizations will use AI to eliminate more than half their middle management. Zuckerberg — who cut 22,000 jobs, mostly managers — put it with his trademark subtlety: “I don’t think you want managers managing managers managing managers managing the people doing the work.”
At one AI-first healthcare company, a 10-person software team became three. One product owner, one engineer, one architect. The seven who left were the execution layer — translating decisions into code. The three who stayed were the ones who knew which decisions to make.
The restructuring puts more weight on the people who know which decisions to make. If you have that expertise, it is worth understanding where it enters the work before deciding what the change means for your career.
What the payroll data shows
Stanford’s Digital Economy Lab analyzed ADP payroll data covering millions — literally millions — of U.S. workers. They measured what actually happened to employment by age group in AI-exposed jobs since late 2022.
Workers aged 22 to 25? Down 16%. Young software developers hit hardest: 20% below their 2022 peak.
Workers over 35 in those exact same jobs? Up 6 to 12%.
Employment moved in opposite directions for the two age groups within the same occupations and period.
Could this simply reflect an economy favoring older workers? In healthcare roles with low AI exposure, all age groups grew equally. The divergence was specific to AI.
HBR research puts a number on it: 17.8 million U.S. workers are in roles where AI automates entry-level tasks while senior roles stay protected. The organizational pyramid is compressing from below. The valuable work is moving up.
Stanford researchers call it the distinction between codified knowledge and tacit knowledge. AI eats the codified — the formal training, the documented procedures, the learnable rules. What it can’t touch is the tacit — the intuition from years of solving real problems, the pattern recognition from a career’s worth of consequences, the judgment that knows not just what the procedure says but when the procedure is wrong.
Your decades of experience aren’t a resume line anymore. They’re a competitive moat. And unlike most moats, this one gets deeper with age.
The value of verification
A recurring problem with AI is that it is spectacularly good at producing things that look right but aren’t. It’s the intern who writes beautiful memos full of made-up statistics. Except this intern works at the speed of light and never looks nervous.
Last year, 47% of enterprise AI users made at least one major decision based on completely fabricated AI output. Nearly half. And these weren’t gullible people — they were professionals in serious environments who trusted output that was confidently wrong.
Deloitte’s 2026 report found that only one in five companies has mature oversight for its AI agents. The tools are multiplying. The ability to check them? Not so much.
This is where experienced professionals become indispensable.
A BCG study with 758 consultants found that lower-skilled performers gained 43% in productivity with AI. Top performers? Only 17%. On the surface, that looks like AI levels the playing field.
It doesn’t.
The researchers discovered what they called the “jagged frontier.” AI performs brilliantly on some tasks and falls on its face on others — and only the experienced consultants could tell which was which. The juniors couldn’t detect when AI was confidently wrong. The experts could. Every time.
That’s the verification economy. AI output is noise until someone with real domain knowledge looks at it and says “yes, this holds” or “no, this will blow up on Tuesday.” And that someone needs years of watching things blow up to know the difference.
Klarna: A Cautionary Tale with a 61% Raise
Between 2022 and 2025, Klarna cut its workforce from 5,527 to 2,907. Their AI chatbot replaced 853 customer service agents. Middle-layer coordination roles evaporated. For a moment, it looked like the playbook of the future.
Then customer satisfaction cratered.
The AI lacked empathy. It failed at nuance. It gave generic, repetitive responses to complex problems. CEO Sebastian Siemiatkowski admitted publicly: “We went too far. We focused too much on efficiency and cost.”
They started rehiring into different roles.
The new model: AI handles the volume, experienced humans handle the judgment. The senior staff who stayed? Their average pay went from $126,000 to $203,000. That’s a 61% raise. Revenue grew 108% even with half the headcount — but only after they brought the judgment layer back.
The execution middle collapsed. The judgment layer became worth more than ever. Klarna spent a few billion dollars learning the value of a distinction: the person who knows whether something is actually right contributes more than the person who only produces it.
”But Experienced People Resist AI”
The strongest objection to everything I’ve written isn’t that experience is worthless. It’s that experienced professionals are often the slowest to adopt new tools. And there’s real data behind this. Research on cognitive entrenchment shows that deep expertise can make people less flexible, not more. Baby boomers report a 35% confidence drop when confronting AI tools. And if experience naturally led to adaptability, incumbent companies wouldn’t keep face-planting during disruption.
I have seen this resistance. A 55-year-old saying “I didn’t need it for 30 years” may have a substantial career behind that judgment. But using experience to defend familiar methods from every change closes off the opportunity to combine that experience with new tools.
This does not mean all experienced professionals will thrive. It’s that experience becomes the scarce differentiator for those who do adopt. PwC’s 2025 barometer found that workers with AI skills command a 56% wage premium. Double what it was in 2024. The premium goes to people who combine domain expertise with AI capability. Not one without the other.
There’s also the “melting ice cube” objection — that AI is compressing decades of knowledge into months. And it’s true that AI diagnostic models already outperform cardiologists on certain structured tasks. For a cardiologist whose work centers on one structured task, that is a real limit to the argument.
But the same Stanford data shows humans outscoring AI two-to-one when given extended time on complex, contextual problems. AI is fast. Humans are deep. The verification role doesn’t require knowing more facts than AI. It requires knowing which facts matter in this context, what’s missing from the output, and what happens downstream when you get it wrong.
The distinction that matters: experience as identity (rigid, defensive, “I’ve always done it this way”) versus experience as inventory (flexible, deployable, “I’ve seen this pattern before and here’s what actually happens”). If you cling to your expertise as proof of status, AI will expose the rigidity. If you treat it as raw material, AI becomes a multiplier.
Where You Sit in the Stack
Think of value in an AI-enabled organization as four layers. AI pressure comes from below, like water rising.
Execution (collapsing). Research, analysis, drafting, coordination. If your job is primarily doing known work according to established procedures, this layer is shrinking fast. Not disappearing — compressing. Fewer people, more AI, same output.
Direction (elevated). Setting intent. Framing problems correctly. Choosing which questions to ask. This requires the pattern recognition that only comes from years of seeing what works and what spectacularly doesn’t. You can’t shortcut it with a prompt.
Verification (critical). Looking at AI output and knowing whether it’s right, useful, or dangerous. Not formatting — substance. Is this analysis sound? Will this recommendation survive contact with reality? This is where domain expertise becomes the quality gate.
Judgment (irreplaceable). Knowing when to override the system. When a technically correct answer is organizationally wrong. When data says one thing but your gut — trained by decades of consequences — says another. This is the layer that compounds with age. AI can’t compress it because it’s built from experiencing things going wrong, and that takes time.
The more AI handles Layer 1, the more valuable everything above it becomes. If you’ve spent 25 years building Layers 2 through 4, those are capabilities to bring into AI-assisted work.
What This Doesn’t Change
Not every experienced professional will benefit. If your expertise is narrow, procedural, and codifiable — if it’s the kind of pattern matching AI already does well — the advantage shrinks. The thesis applies to tacit knowledge: judgment, context, relationships, political awareness. The stuff you can’t write into a manual.
AI will also keep getting better at verification. Routine checks are already being automated. If you’re staking your career on catching typos or checking compliance boxes, that’s an execution-layer role wearing a judgment disguise.
And none of this is automatic. The 56% wage premium goes to people who actually use the tools. The advantage depends on combining experience with practical use of the tools.
Try it on work you know
Try an AI tool this week on a task you know well. Ask it to draft something in your domain — a report, an analysis, a recommendation. Then look at the output with your 20 years of experience and notice what it gets wrong. The subtle errors. The things that look polished but aren’t quite right. The recommendations that would work in a textbook but blow up in your specific context.
That feeling — where you see through the confident output to the flaw underneath — that’s the thing AI can’t do. That’s your advantage. And unlike practically everything else in this economy, it gets more valuable with age.
The execution middle is compressing. The judgment layer is expanding. If you’ve spent decades building domain expertise, this isn’t a threat.
It’s the moment your experience was built for.
You can begin with a piece of work you already understand, and use that understanding to examine what comes back. Ya tardáis.
Sources
- Korn Ferry survey on disappearing middle managers — success.com
- Gartner prediction: AI to eliminate 50%+ of middle management by 2026 — fortune.com
- Stanford Digital Economy Lab, ADP payroll analysis by age group — time.com
- HBR: How Gen AI could change the value of expertise — hbr.org
- 47% of enterprise AI users acted on fabricated output — guildhawk.com
- Deloitte 2026 State of AI in the Enterprise — deloitte.com
- BCG study: 758 consultants and the “jagged frontier” — hbr.org
- Klarna workforce cuts and rehiring — fortune.com
- Klarna CEO admits overreach — lasoft.org
- PwC 2025 AI Jobs Barometer: 56% wage premium — pwc.com