# AI Deskilling: The Skill Your Team Loses When AI Helps > A study of experienced doctors found their unaided skill dipped after months of AI help. What AI deskilling means for Indian teams and how to prevent it. - Source: https://bodhih.com/ai-deskilling-skills-fade-when-ai-does-the-work-india-teams-training/ - Publisher: Bodhih Training (Bengaluru, India; since 2008) - Last updated: 2026-10-16 - Contact: solutions@bodhih.com · +91 99000 11601 --- Picture an experienced doctor in Poland, more than two thousand procedures behind her. She is excellent at spotting the faint, flat growths that can turn into cancer. Then her clinic installs an AI assistant that highlights suspicious areas on screen. Detection goes up when the assistant is on. Everyone is pleased. A study of experienced doctors found their unaided skill dipped after months of AI help. What AI deskilling means for Indian teams and how to prevent it. Key takeaways - It was an observational study, not a randomised trial, so it shows an association rather than proof of cause. Even so, it is the most-cited hard evidence for "AI deskilling" and has pushed the idea into boardrooms. - Most other evidence so far comes from medicine, coding and education, and much of it is preliminary. There is little rigorous data on ordinary office work, so the right posture is prepare, not panic. - The risk is real wherever AI takes over the practice of a skill. Writing, analysis, first-draft thinking, customer judgement and code review all qualify. - The fix is not "use less AI". It is to design where people still practise, check and think, and to measure it. Last updated 16 Oct 2026 · Bodhih Insights team · reviewed by [Siddharthan VGJ](https://bodhih.com/siddharthan-vgj/), Founder Months later, researchers look at what happens on the days the assistant is *off*. The doctor is now a little worse than she was before the tool arrived. Nothing about her training changed. Nobody told her to relax. The skill simply faded, quietly, because the machine was doing part of the looking. That finding, from a real study, is the cleanest warning yet for every company rolling out AI tools: **a team can get faster and weaker at the same time.** ## The Pattern: Output Rises, Ability Quietly Falls We usually measure AI adoption by output: reports produced, tickets closed, emails sent per hour. All of those can climb while the underlying ability sinks, because output with AI and ability without AI are two different things. Think of a driver who uses satnav for every trip for a year. Trips get faster and fewer wrong turns happen. Take the phone away and they cannot find their way around their own city. Nothing was wrong with the satnav. The skill of building a mental map just never got used. Work has plenty of mental maps: - The analyst who has not built a model from a blank sheet in months. - The salesperson who has stopped preparing for a call because the tool "summarises the account". - The manager who now drafts every piece of feedback with a prompt and has forgotten how to start a hard conversation. - The junior engineer who ships code that works and cannot explain why. Our earlier posts on [workslop](https://bodhih.com/workslop-ai-generated-work-india-teams-training/) and [AI brain fry](https://bodhih.com/ai-brain-fry-mental-fatigue-ai-tools-india-teams-training/) described what AI does to the quality of work and to the people reviewing it. Deskilling is the slower, longer-term cousin: what AI does to the *person*. ## Why the Junior Pipeline Is the Real Exposure Experienced professionals have a deep reservoir to draw on. Even when some of it fades, they still know what good looks like. The doctor in the study dipped by six points; she did not forget how to do the job. The sharper risk lies with people who never built the reservoir. Reports from Indian hiring markets suggest that entry-level tech hiring is already under pressure, with an ICRIER study of 650 IT firms finding a moderation in hiring concentrated at the entry level ([People Matters](https://www.peoplematters.in/news/workforce-planning/fresher-hiring-intent-rises-to-75percent-but-it-demand-loses-ground-report-51352)). Tasks that used to belong to freshers, such as basic testing, documentation and first drafts, are the very tasks AI now takes first. Those tasks were never only about the output. They were the apprenticeship. They were how a fresher learned what an error looks like, how a client thinks and how a good document is structured. If the practice rungs are removed, you get a workforce that can *direct* AI but cannot *judge* its output, because they never did the work by hand. That is the uncomfortable question for every L&D head: **if AI does the learning-by-doing, where will your next generation of experts come from?** ## Four Signs Your Team May Be Deskilling None of these is proof. Together they are a reason to look closer. 1. **Fluent, but fragile.** People produce confident work with AI and struggle to explain the reasoning or reproduce it without the tool. 2. **Review has become a skim.** Checking AI output is turning into approving AI output. 3. **Stalling on blank pages.** Tasks that require starting from nothing, such as a new client situation or an unusual problem, now feel disproportionately hard. 4. **Rising dependence in your best-performing group.** The first people to adopt AI heavily often show the effect first. This is related to what we described in [the two-speed workforce](https://bodhih.com/two-speed-workforce-self-taught-ai-skills-india-teams-training/). ## What to Do: Keep the Muscle While Using the Machine Pilots do not stop using autopilot. They do keep practising manual flying, because the day the system fails is the day skill matters. Companies can borrow the same logic. 1. **Decide which skills are "load-bearing".** For each role, name the three or four capabilities that, if lost, would make the person unable to supervise AI or work through a failure. Protect these first. 2. **Build "AI-off" practice into the week.** A first draft by hand on one task per week, a case solved from scratch in a team meeting, a quarterly exercise with no tools. Treat it as a gym session, not a punishment. 3. **Rebuild the apprenticeship deliberately.** If AI takes the junior tasks, create structured substitutes: shadowing, supervised case work, "explain your answer" sessions and rotations. Our [campus-to-corporate training](https://bodhih.com/campus-to-corporate-training/) is designed around exactly this problem. 4. **Make people explain, not just deliver.** Ask "how would you check this?" and "what would change your mind?" as standard questions in reviews. 5. **Measure ability, not only output.** Productivity dashboards will not catch deskilling. Periodic skill assessments done *without* AI will. See [skill gap analysis](https://bodhih.com/skill-gap-analysis/) for how to begin. 6. **Train the checkers.** Reviewing AI work is a skill in itself and it also fades. It needs deliberate practice and standards. ## How Bodhih Helps Bodhih designs programs around the idea that AI skills and human skills grow together. - [AI Literacy Training](https://bodhih.com/ai-literacy-training/) teaches what AI does well, where it fails and how to verify its output, so people use it with their eyes open. - [Critical Thinking Training](https://bodhih.com/critical-thinking-training/) builds the habit of questioning a fluent answer, which is the first thing to fade and the first line of defence. - [Leading Human + AI Teams Training](https://bodhih.com/leading-human-ai-teams-training/) helps managers decide which work stays human, set review standards and protect skill growth in their teams. - [Campus to Corporate Training](https://bodhih.com/campus-to-corporate-training/) rebuilds the early-career learning curve with structured practice instead of leaving it to chance. - [Coaching for Managers](https://bodhih.com/coaching-for-managers/) equips leaders to run "explain your thinking" conversations that keep reasoning visible. - [Assessments](https://bodhih.com/assessments/) give you evidence of real skill levels, before and after training, so you can see whether capability is rising or just output. - For trainers building these programs, the [AI for Trainers course](https://bodhih.com/ai-for-trainers-course/) shows how to use AI to design learning without hollowing out the practice that makes learning stick. Corporate batches are private and tailored to your roles and tools. Open workshops are fixed-date public batches. ## The Bigger Idea Every tool in history has changed which skills matter. Calculators did not end mathematics, but they did end a certain kind of mental arithmetic, and most of us accepted the trade. The difference with AI is the range. It can take over *thinking* tasks, not just calculating ones, and it can do so inside the same afternoon that someone is learning them. So the real choice for a leader is not "adopt AI or not". It is: **which skills do we want to be able to do without the machine, and what are we doing this quarter to keep them alive?** Teams that answer that early will have people who use AI as an amplifier. Teams that skip it may find, one day when the tool is down, wrong or simply asked something new, that nobody remembers how to start. ## Frequently asked questions ### What is AI deskilling? AI deskilling is the gradual loss of a skill because AI does the practice that used to build and maintain it. Output may stay high or even rise while a person's *unaided* ability slips, which makes it hard to notice using normal productivity measures. ### Is there real evidence that AI makes people worse at their jobs? There is early evidence, mainly from medicine. A 2025 observational study in *The Lancet Gastroenterology & Hepatology* found that experienced endoscopists' unaided detection fell by about six percentage points after AI assistance was introduced at four Polish centres ([study record](https://wrap.warwick.ac.uk/id/eprint/191005)). Evidence for ordinary office work is limited and mostly preliminary, so treat it as a risk to manage, not a settled fact. ### Does this mean companies should use less AI? No. AI delivers real gains, and the study itself noted that AI can improve detection while it is switched on. The aim is to use AI deliberately: keep practising the load-bearing skills, protect early-career learning and check ability regularly. ### Who is most at risk of skill loss from AI? Freshers and early-career employees, because the entry-level tasks AI absorbs first are the ones that used to teach them the craft. Heavy daily users in any role can also be affected if they stop doing tasks unaided. ### How can we measure whether our team is deskilling? Compare output measures with ability measures. Run short assessments or practical exercises *without* AI at regular intervals, and ask people to explain their reasoning on past work. A fall in unaided performance while output rises is the warning sign. Bodhih's [assessments](https://bodhih.com/assessments/) can help structure this. ### What training helps prevent AI deskilling? A mix works best: [AI Literacy Training](https://bodhih.com/ai-literacy-training/) for everyone, [Critical Thinking Training](https://bodhih.com/critical-thinking-training/) to keep judgement sharp, [Leading Human + AI Teams Training](https://bodhih.com/leading-human-ai-teams-training/) for managers, and [Campus to Corporate Training](https://bodhih.com/campus-to-corporate-training/) for early-career hires. ### Where do the facts in this article come from? The colonoscopy findings are from Budzyń et al., *The Lancet Gastroenterology & Hepatology*, 2025 (volume 10, pages 896 to 903), an observational study of four centres in Poland. The India hiring context is from [People Matters](https://www.peoplematters.in/news/workforce-planning/fresher-hiring-intent-rises-to-75percent-but-it-demand-loses-ground-report-51352), reporting an ICRIER study. The study findings are observational, so use them as directional.