By Zaahid Allie, Data Analytics Consultant at Insight Consulting.
Part 1 of 3: Cognitive dystrophy in the age of agentic AI
I have used the wrong word in the title of this series, and I want to explain why it is the right one.
The literature calls what I am describing cognitive atrophy, or cognitive debt, or deskilling. Atrophy is what happens to a muscle that stops being used: it was built, and now it is wasting. That describes most of what follows, and it is real.
But dystrophy is a different thing. Dystrophy is when the mechanism that builds and maintains the tissue is itself faulty, so it never develops properly in the first place. That is the version almost nobody is discussing, even though it is the harder problem for a business.
The tasks AI absorbs first are disproportionately the tasks people learn on. Reconciling a figure by hand. Chasing why a chart disagrees with a report. Rewriting a script five times until it is right. Working out what a request actually means before answering it. These tasks are tedious, and they are also how people are built. Automate them without deliberately replacing them, and you do not get a faster junior. You get someone who never becomes a senior.
Four mechanisms:
You will recognise at least two of these from your own life before you recognise them in your work.
- Offloading. How many phone numbers do you know by heart? Twenty years ago, most people knew dozens. Now it is two, and one of them is your own. Nothing went wrong here: the numbers are safer in the phone than they ever were in our heads. But the capability is gone, and we did not decide to give it up.
- Effort substitution. People who have followed GPS around the same city for ten years often still cannot describe how that city fits together. People who used to get lost there can. The difference is not intelligence or attention; one group had to build a map, and the other never needed one. Learning is largely a byproduct of effort, so removing the effort removes the learning with it.
- Automation bias. Autocorrect is the everyday version of this. Everyone reading this has sent a message with a confidently wrong word in it, because the suggestion looked right and was never checked. When a system offers you a recommendation, you under-verify it. AI makes this considerably worse, because how fluent an answer sounds tells you nothing about whether it is correct, and fluency defeats the quick plausibility check most of us actually rely on.
- Deskilling. This one has hard numbers, and they come from medicine rather than knowledge work. Nineteen experienced specialists across four hospitals, each of whom had performed more than two thousand colonoscopies, were studied before and after an AI tool was introduced to help spot polyps. Their detection rate when working without the AI fell from 28.4% to 22.4% within months: a 20% relative drop, in senior people, in the core skill of their profession (Budzyn et al., The Lancet Gastroenterology & Hepatology, August 2025).
That study matters here for a simple reason: these were not novices. They were experts, and the skill still slipped, quietly, in months, while the tool made them look just as capable as ever.
Why this is a business problem, not a personal one
It would be comforting to treat this as an individual discipline issue: some people will keep their edge, others won’t, and that is their own affair. It isn’t, quite. The mechanisms above do not ask permission. Offloading, effort substitution, and automation bias operate on everyone who uses the tool, seniority and good intentions included. And the fourth one, deskilling, is precisely what shows up on a balance sheet a year or two later, when the person everyone relied on to catch the odd mistake can no longer catch it, because the everyday practice that built that instinct quietly stopped happening.
None of this is an argument against using AI. I use it constantly, and I will get to exactly why that is compatible with everything above. But the first step is naming the thing accurately. Atrophy is a skill going soft from disuse. Dystrophy is a skill that never gets the chance to form. Most conversations about AI and skills are having the atrophy conversation. The one that should worry a business more is the dystrophy conversation, because you cannot rebuild a muscle that was never built.
In the next post, I want to give you a frame for thinking about which parts of the work are safe to hand over completely, and which parts need to stay yours on purpose. It comes, unsurprisingly, from a machine we all trust rather more than we should: the car.
This is Part 1 of a three-part series on cognitive dystrophy and agentic AI. Part 2 covers the car framework and how to draw your own 80/20 line; Part 3 covers vibe coding, agentic AI, and a practical checklist for Monday morning.
Sources referenced in this post:
- Budzyn et al., “Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy”, The Lancet Gastroenterology & Hepatology, August 2025” multicentre observational study, 19 experienced specialists.
This article first appeared in Insight Edge, our newsletter on LinkedIn.