De-skilling and loss of agency are real harms, but only at the level of a person and a task. Zoom out to a workforce or a society, and the same words stop being accurate.
I have been building a working vocabulary for what AI does to the people who use it. Automation bias, complacency, de-skilling, loss of agency. The literature is rich, and most of it is sound.
But two of those terms kept bothering me. Every time I said "de-skilling" or "loss of agency" in a broad sense, I felt I was only describing half the coin. Something was being lost, yes. Something else was being gained, and the word hid it.
Here is where I have landed: the harm is local, the change is global. We need different words for each.
Two levels, two vocabularies
The micro level is one person, one task, one job. The macro level is a workforce, an industry, an economy. The same shift looks very different from each.
| Micro: a person and a task | Macro: a workforce or society | |
|---|---|---|
| Skill | De-skilling: a capability you needed and lost | Re-skilling: capability transformed, old skills traded for new |
| Agency | Loss of agency: a decision you should still own, now made for you | Agency redistribution: who decides what is shifting, not simply shrinking |
| Right question | Did we lose something we needed? | Are we steering the change, or drifting? |
At the micro level, "loss" is precise. At the macro level, it is a judgment disguised as a description.
Micro de-skilling is real
None of this softens the local harm. When a person loses a skill they still need, that is a failure, full stop.
Aviation learned this decades ago. Pilots who follow the automated flight path lose the habit of hand-flying, and the skill is missing on the one day the automation hands control back. Lisanne Bainbridge called this the irony of automation in 1983: automate the easy parts, and the human is left with the hard parts, using skills they no longer practise.
Medicine is now showing the same pattern with AI. In one 2025 study, endoscopists who had worked with AI assistance detected fewer adenomas when working unaided, dropping from 28.4% to 22.4%.
AI adds a twist earlier automation did not. It can erode the very skills you need to check the AI. A quality reviewer who stops writing investigations eventually loses the ability to tell a good one from a plausible one. That is the de-skilling I worry about most.
Macro re-skilling is the bigger story
Now zoom out. Did the industrial revolution de-skill humanity? It wiped out whole trades and the skills that went with them. Very few of us can shoe a horse or weave cloth by hand. Yet nobody would describe the last 200 years as humanity losing capability.
Skills were traded, not simply lost. That is what happens at scale, and AI will be no different.
So at the macro level I don't say "de-skilling is bad." I say capability is being transformed, and the real risk is not the transformation itself. The real risk is letting it happen by drift instead of by decision.
The question for a leader is never "how do we stop de-skilling?" You can't, and you shouldn't try. The question is: which capabilities are we deliberately trading away, and which ones must we keep?
Agency follows the same pattern
"Loss of agency" is the phrase I hear most often, and the one I trust least.
I delegate to AI constantly. In most technical, knowledge-heavy work it is already better than I am. It still makes mistakes, but fewer than I would, across far more ground. Handing it that work doesn't make me less of an agent. Often it frees me to direct more.
Philosophers make the same point: choosing to delegate, freely and with your eyes open, is itself an exercise of autonomy. Delegation is not loss.
Agency erodes locally when delegation takes away your ability to choose, understand, redirect or refuse. If you can still set the direction, question the trade-offs and pull the work back, you haven't lost agency. You've extended it.
At the macro level, the honest phrase is agency redistribution. Decisions are moving between people, institutions and machines. Some people gain reach they never had; some lose a say they used to have. Calling all of it "loss" misses half of what is happening, and it makes it harder to see where the real losses are.
What people actually fear losing is often significance
When I pressed myself on why "loss of agency" felt so heavy, I realized I was blending two things. Agency is the power to choose. Significance is the sense that what you contribute matters.
For most of history, people have earned their bread by the sweat of their brow, and we have measured our worth largely by what we contribute. If AI takes over much of that work, we may gain a kind of freedom humanity has never had at scale. That is more agency, not less.
But it can still wound our sense of significance. A person can keep every choice and still lose the feeling of being needed.
That deserves its own name, something like role or significance displacement, and its own attention. Treating it as an agency problem sends us looking for the wrong fix.
Govern the transfer, don't fight it
If the harm is local and the change is global, the leader's job is to govern the transfer on purpose. In practice, that means asking four questions about every AI-assisted process:
- Which skills must stay in human hands? Especially the ones needed to check the AI, recover when it fails, or take over in an emergency.
- Which skills are we choosing to retire? Name them. A skill you retire deliberately is a trade; a skill you lose without noticing is drift.
- Can a person still choose, understand, redirect and refuse? If not, you have crossed from delegation into local loss of agency.
- Does the role still give people something that matters? If not, expect disengagement, even where skills and choices are intact.
In regulated industries, this isn't philosophy. It's what an auditor will ask when the AI is wrong and a human signed off.
Diagnose locally, describe globally
The vocabulary we use shapes the decisions we make. If every change is "loss," we either panic or tune it out.
So I'm adopting a simple discipline. At the level of a person and a task, I'll name the harm plainly: de-skilling, loss of agency. At the level of an organization or a society, I'll describe the change honestly: re-skilling, agency redistribution. And I'll keep significance as its own question.
Where do you see this in your own work? Which skill or decision would you refuse to hand to AI, even if it did the job better?
First published in my LinkedIn newsletter, AI in Regulated Industries.
Jared Clark
Principal Consultant, Certify Consulting
Jared Clark is the founder of Certify Consulting, helping organizations achieve and maintain compliance with international standards and regulatory requirements.