Deskilling is the loss of skill in a workforce or an individual when technology or work design removes the practice that maintained it. The term is roughly fifty years old, it predates AI by decades, and almost everything currently being said about AI and capability was said first about the assembly line, the autopilot and the calculator.
The answer, in one line
The loss of skill in a workforce or an individual when technology or work design removes the practice that maintained it. The term originates with Harry Braverman in 1974, who argued that industrial management separated the conception of work from its execution.
Definition#
Deskilling: the reduction of skill required or retained in a role, caused by the transfer of skilled elements of the work to a machine, a procedure or another group of workers. It can affect what a job demands, what a person can still do, or both, and those are worth distinguishing.
Where the term comes from#
Harry Braverman, 1974. Labor and Monopoly Capital argued that industrial management systematically separated the conception of work from its execution, concentrating knowledge in management and leaving execution progressively less skilled. It is a contested thesis and it generated decades of argument, including strong evidence that technology also upskills in many settings. It is the origin of the word as a term of art.
Lisanne Bainbridge, 1983. "Ironies of Automation" made the sharper operational point: automate the routine parts of a task and you leave the human with the hardest residue, monitoring and exception handling, while removing the routine practice that built the competence to do it. The irony is that automation makes the remaining human role harder, not easier, precisely as it erodes the skill needed for it.
Everything since is elaboration. Aviation produced decades of research on manual flying skill under autopilot. Medicine has produced the sharpest recent evidence.
The strongest current evidence#
Budzyń and colleagues, publishing in Lancet Gastroenterology and Hepatology in 2025, examined colonoscopies at four Polish centres before and after AI polyp-detection tools were introduced. Adenoma detection in unassisted colonoscopy fell from 28.4 per cent to 22.4 per cent, a drop of 6.0 percentage points.
This is the first real-world clinical evidence of the effect, and adenoma detection rate is a recognised clinical quality indicator rather than a laboratory task. It is still a proxy for skill and not skill itself, which is how the graded entry states it. It is also retrospective and observational rather than randomised, so change over time from other causes cannot be excluded. A consequential finding on a design that cannot yet prove causation.
Outside medicine, the closest long-run analogues are memory and navigation rather than reasoning. Habitual satnav users show worse unaided spatial memory, with steeper decline over three years of heavier use.
Three distinctions that get collapsed#
Deskilling the job versus deskilling the person. A role can require less skill while the people in it retain theirs, and a role can look unchanged while the people in it lose capability. Only the second is measured by asking what someone can do unaided.
Deskilling versus skill substitution. Losing arithmetic to the calculator while gaining modelling is a trade, not a loss. Whether a given case is a trade or a loss depends on whether the lost skill was load-bearing for judgement you still need.
Deskilling versus never-skilling. An experienced person losing a skill and a new entrant never acquiring it look identical on a capability audit and require completely different responses. The second is what this research calls the missing rungs.
Medicine has since given the argument a three-way vocabulary#
Ke and sixteen colleagues, writing in Nature Medicine on 22 May 2026, separate three failures that the single word deskilling had been carrying. Deskilling is the decay of a competence a person once held. Never-skilling is the failure to form that competence at all during training, because AI substituted for the effort that would have built it. Mis-skilling is competence formed wrongly, shaped around what the system does rather than around the work.
The distinction matters for what you would do about it. Deskilling has a capability to restore; never-skilling has none, so the remedy is curricular rather than remedial. The three terms belong to those authors and are not SuperSkills coinages. The paper is also a Perspective and reports no new data, and the authors state in their own words that direct causal evidence linking AI exposure during training to competency failure in medical trainees does not exist. Cite it for the taxonomy and never as evidence of harm. Graded entry.
The version that shows up in identity before it shows up in a metric#
Ehsan and colleagues spent twelve months inside a five-site North American hospital group through the first year of routine use of an AI treatment-planning system, with 42 participants across radiation oncology. Planning cycles shortened by roughly 15 per cent and confidence rose. By month nine, several dosimetrists said their unaided proficiency had worsened over the year. The authors call the first effect the one the organisation measured and the second asymptomatic, because every dashboard showed only the improvement. Their term for the mechanism is intuition rust: expert judgement dulling underneath output that still looks correct.
Two things about this study are routinely got wrong in circulation, and this page corrects them instead of passing them on. The participants plan radiotherapy treatment; they are not radiologists reading images. And the system is optimisation-based, not generative. It is qualitative work, the skill claims are self-reported, and it cannot be set beside the Polish colonoscopy data as a second measured result. What it gives that nothing else here does is the account of what deskilling feels like from inside a profession while the numbers are still good. Graded entry.
What is not established#
Whether generative AI produces durable deskilling in cognitive work is not established. The clinical evidence is one observational study; the education evidence is one field experiment showing a 17 per cent drop when access was withdrawn; neither is replicated. The historical literature is genuinely mixed, with substantial evidence of upskilling alongside the deskilling cases. Anyone stating this confidently in either direction is going beyond the evidence, including anyone arguing it from this site.
A measurable question instead#
The most useful move is to stop asking whether AI deskills and start asking a measurable question: what can this person or organisation still do without the system, and when did anyone last check? Deskilling is invisible while the tool is present, because performance is fine. It is only observable in the counterfactual, and almost nobody runs it.
That is the argument for treating capability as a stock that depreciates rather than an asset that sits still, which is what capability debt describes at organisational scale.
23 September 2026: skill erosion as 8,800 employees reported it, and what a self-report can show#
The largest survey yet to ask employees the question directly was published on 21 September 2026. IBM’s Institute for Business Value, with Oxford Economics, put the same questions to 1,500 chief human resources officers and 8,800 full-time employees in 28 countries between April and June, and its 2026 CHRO Study reports that 60 per cent of employees worry AI is eroding their skills, that three quarters of those say it already has, and that 46 per cent of executives name skill erosion as a top concern, as Campus Technology reported the figures. The study separates skill erosion, existing skills that degrade as AI takes over the underlying work, from the skills gap, the new skills people need to use the tools, and notes that the 80 per cent of organisations with a reskilling roadmap are addressing the second. That is the first of the three distinctions above in a vendor’s vocabulary: a roadmap for what people must learn says nothing about what they are ceasing to be able to do. Two cautions. Every figure is self-reported in a commercial study by a company that sells AI to the organisations surveyed, and a worry is not a measurement; the endoscopists and the dosimetrists above were measured, and the 8,800 were asked. And the finding cuts against one of this page’s own claims, that deskilling is not reported by the deskilled because performance stays fine while the tool is present. Six in ten reporting the worry suggests people can feel the practice going even when their output does not show it, which is what the dosimetrists said at month nine. Whether the feeling tracks the loss is the study nobody has run, and the measurable question at the end of this page is still the one to ask.
Related SuperSkills research#
On the organisational version, capability debt. On new entrants rather than incumbents, the missing rungs and the missed reps. On the cognitive mechanism, cognitive offloading. On the counter-argument that this is a revaluation of skill and not a loss of it, is deskilling real, or a rescaling. On what the evidence does and does not establish, what we actually know.
Key sources
- Budzyń, K. et al. (2025). Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy. Lancet Gastroenterology and Hepatology.
- Bainbridge, L. (1983). Ironies of Automation. Automatica, 19(6).
- Dahmani, L. and Bohbot, V. D. (2020). Habitual use of GPS negatively impacts spatial memory during self-guided navigation. Scientific Reports, 10.
- Bastani, H. et al. (2025). Generative AI can harm learning. PNAS.
- Ke, Y. et al. (2026). AI-induced never-skilling in medical education. Nature Medicine, 32(6), 22 May 2026.
- Ehsan, U. et al. (2026). From Future of Work to Future of Workers. CHI '26, ACM.
- IBM Institute for Business Value (2026). 2026 CHRO Study: Designing the Thinking Organization. IBM, with Oxford Economics, 21 September 2026. Graded entry.
About this definition#
Deskilling is an established term from the sociology of work, originating with Braverman in 1974, and is not a SuperSkills coinage. Findings are attributed to the studies that produced them and kept separate from the interpretation.
About this research#
Written by Rahim Hirji, author of SuperSkills (Kogan Page, 2026), keynote speaker on AI and human capability, and founder of The SuperSkills Intelligence Company.
Explainer · SS-2026-079 · Graded against the published rubric · 1 peer-reviewed study and 1 institutional survey
Hirji, R. (2026). What is deskilling?. The SuperSkills evidence base, SS-2026-079. https://thesuperskills.com/research/what-is-deskilling. Last reviewed 23 September 2026.
An evidence review by Rahim Hirji, not peer-reviewed research. For a material claim, cite the underlying study as well; every study here carries its own permanent link.
How citations and IDs work