Something uncomfortable is happening in offices across America, and most managers haven’t noticed yet. The same artificial intelligence tools companies deployed to make workers faster and more productive are systematically hollowing out the very expertise those workers were hired for. It’s not a glitch. It’s the design working as intended.
A growing body of research and firsthand accounts from workers suggests that heavy reliance on AI assistants — from code generators to writing tools to data analysis platforms — is producing a workforce that’s increasingly dependent on machines for tasks they once handled with confidence and skill. The phenomenon has a name now: AI deskilling. And it’s becoming one of the most consequential workforce issues of the decade.
Business Insider reported on the growing alarm among researchers and employers about deskilling, noting that workers who lean on AI for core job functions are experiencing measurable declines in their ability to perform those functions independently. The pattern is consistent across industries — software engineering, financial analysis, legal research, medical diagnostics, content creation. Workers get faster. Then they get worse.
The mechanism isn’t mysterious. When a software developer uses an AI coding assistant to generate functions they once wrote by hand, the repetition that builds and maintains skill disappears. The developer still ships code. The code might even be better in the short term. But the developer’s own understanding of what’s happening under the hood atrophies. Slowly at first. Then not so slowly.
The Productivity Trap
Here’s the paradox that makes deskilling so insidious: by every metric companies typically track, AI-assisted workers look more productive. They produce more output in less time. They hit deadlines. They handle larger workloads. Quarterly numbers go up.
But productivity metrics don’t capture what’s being lost underneath. They don’t measure whether a financial analyst still understands the assumptions behind the model an AI just built for her. They don’t flag that a junior lawyer can no longer construct a legal argument from scratch because he’s been summarizing AI-generated briefs for two years. The dashboards stay green while the foundation erodes.
Research from Microsoft and Carnegie Mellon, cited by multiple outlets covering the deskilling phenomenon, found that developers using GitHub Copilot wrote code roughly 55% faster. Impressive. But follow-up studies began revealing that those same developers showed declining performance on coding assessments that required independent problem-solving. Speed up, skill down.
This isn’t an abstract concern. Companies are already encountering the consequences. When AI systems go down — and they do — teams that have been heavily dependent on them struggle to maintain baseline operations. When novel problems arise that fall outside an AI model’s training data, workers who once would have improvised solutions find themselves stuck. Waiting for the tool to come back online. Or escalating to the shrinking number of colleagues who still know how to do things the old way.
The problem compounds over time. Junior employees are most vulnerable because they’re building foundational skills — or they should be. A first-year consultant who uses AI to generate every client presentation never develops the analytical thinking that separates a competent consultant from a great one. The AI handles the how. Nobody learns the why.
And companies are hiring for this new reality without fully understanding what they’re giving up. Job postings increasingly emphasize “AI fluency” and “prompt engineering” over domain expertise. The implicit message: you don’t need to know the subject deeply. You just need to know how to ask the machine.
That’s a bet. A big one.
What the Research Actually Shows
The academic literature on skill degradation from automation isn’t new. Researchers have studied what happens when pilots rely too heavily on autopilot systems for decades — the answer is that manual flying skills deteriorate measurably, sometimes with catastrophic results. The same dynamics have been documented in manufacturing, where operators who monitored automated systems lost the ability to intervene effectively when those systems failed.
What’s different about the current AI wave is its scope. Previous automation deskilling affected specific roles in specific industries. Large language models and generative AI tools are touching virtually every knowledge-worker function simultaneously. Writing. Analysis. Design. Strategy. Research. Code. The breadth is unprecedented.
A January 2025 study published in Nature Human Behaviour examined how AI assistance affected learning outcomes among participants completing complex analytical tasks. The findings were stark: participants who received AI help during training performed significantly worse on subsequent independent assessments compared to those who struggled through the tasks without assistance. The AI didn’t just help them complete the task — it prevented them from learning how to do it.
So the question isn’t whether deskilling is real. It is. The question is what to do about it.
Some companies are beginning to experiment with what might be called “deliberate difficulty” — intentionally limiting AI tool access during training periods and for certain categories of work. McKinsey, according to reporting from several business publications, has implemented structured programs where junior consultants must demonstrate independent analytical capability before being granted full access to the firm’s AI tools. The firm treats AI access as something earned, not assumed.
Others are taking a different approach entirely. Shopify CEO Tobi Lütke made waves in early 2025 when he told employees that demonstrating a task couldn’t be done by AI would become a prerequisite before any new headcount requests would be approved. The message was clear: AI first, humans where necessary. It’s an efficiency-maximizing posture, but it accelerates the very dependency that deskilling researchers warn about.
The tension between these two approaches — protecting human skill development versus maximizing AI-driven efficiency — is becoming one of the defining management challenges of the moment. There’s no consensus. Not even close.
What’s emerging instead is a split. Companies focused on short-term output metrics are pushing deeper into AI dependency. Companies worried about long-term organizational resilience are trying to maintain human capability even when it’s slower and more expensive. Both camps think the other is making a mistake.
Who Pays the Price
The distributional effects of deskilling deserve more attention than they’re getting. Senior workers who built their expertise before AI tools became ubiquitous retain that knowledge — for now. They can use AI as a genuine augmentation layer because they understand the domain well enough to evaluate, correct, and direct AI output. For them, the tools are additive.
Junior workers don’t have that foundation. They’re building their careers on top of AI from day one. And if the research holds, many of them are building on sand. When the tools change — and they will, rapidly — workers without underlying domain expertise will find themselves stranded. Skilled at prompting a system that no longer exists. Unskilled at the work itself.
This creates a looming generational divide in the workforce. The senior people who can function without AI become more valuable. The junior people who can’t become more replaceable. It’s the opposite of what most companies expected when they rolled out AI tools with promises of democratizing expertise and flattening hierarchies.
There’s a labor market dimension too. Workers who’ve spent years building skills through AI-assisted shortcuts may discover those skills don’t transfer when they change jobs, industries, or when their employer switches AI platforms. The portability of expertise — one of the most valuable things a professional can have — gets undermined.
I’ve been watching technology reshape work since I first got my hands on a computer growing up in the Midwest. Every wave of new tools brings genuine gains and genuine losses. The gains are always obvious and immediate. The losses are always subtle and delayed. That’s what makes them dangerous.
The honest answer is that nobody knows exactly where the deskilling curve flattens out. Maybe AI tools will improve to the point where independent human skill in many domains genuinely doesn’t matter — where the tool is so reliable, so comprehensive, that the human’s role is purely supervisory. Maybe. But we’re nowhere near that point today, and betting your workforce strategy on a hypothetical future capability is a risky move.
What’s clear right now is that the companies treating AI adoption as a simple productivity story are missing something fundamental. The tools work. They make people faster. They also make people more fragile. And fragility, in a business environment defined by disruption and rapid change, is a cost that doesn’t show up on any quarterly report — until it’s too late to fix.
The smartest organizations will figure out how to get the speed benefits of AI without sacrificing the deep human expertise that makes those benefits sustainable. That’s not a technology problem. It’s a management problem. And it’s one that most leaders haven’t even started solving.
WebProNews is an iEntry Publication