Is AI Making Your Team Worse at Their Jobs? Deskilling, and How to Avoid It

AI can make a team worse at the skills it stops practising, a pattern called deskilling. Early research shows it in doctors, developers and knowledge workers, mostly where people hand over whole tasks. It is avoidable: use AI to understand rather than to replace thinking, keep some work done by hand, and check skills directly.

What is AI deskilling?

AI deskilling is the gradual loss of a skill because AI now does that part of the work. Writers who never draft lose fluency. Analysts who never build a model lose the feel for when numbers look wrong. Output may stay good while AI is available, then drop when it is not, or when it errs.

The idea is older than AI. Pilots on autopilot, drivers using navigation apps and accountants with spreadsheets have all faced versions of it. What is new is the range: generative AI touches writing, analysis, research and judgement at once.

The risk is not that people forget everything. It is that the specific skill needed to catch the AI's errors fades, while confidence in the output grows. The guide to plausible AI output covers why those errors are hard to see in the first place.

What does the research say about AI and skills?

The evidence is early but consistent. One medical study found doctors detected fewer growths after months of AI assistance when the AI was switched off. A randomised trial found developers learned less when they delegated coding to AI. Separately, a large survey linked trust in AI with less critical thinking. None is the final word; together they justify caution.

According to the Lancet Gastroenterology & Hepatology study, across four Polish centres, the adenoma detection rate in standard colonoscopies fell from 28.4% before AI exposure to 22.4% after. It was observational and short-term, so it is a signal rather than proof.

According to Anthropic's January 2026 trial of 52 mostly junior engineers learning a new library, the AI group averaged 50% on a later quiz, against 67% for those coding by hand. The largest gap was on debugging. Crucially, people who asked AI conceptual questions scored far better than those who handed it the whole task.

The Microsoft Research and Carnegie Mellon survey of 319 knowledge workers, presented at CHI 2025, found "higher confidence in GenAI is associated with less critical thinking".

Which skills are most at risk?

The skills at risk are the ones AI does for people most often: first drafting, structuring an argument, debugging, spotting an odd number, and research judgement. The cruel twist is that checking AI output needs exactly these skills. A team that loses them becomes worse at catching AI mistakes just as it relies on AI more.

Why are junior staff most exposed?

Because the groundwork AI now does was their training. Juniors learned by drafting the report, building the spreadsheet and writing the first proposal, then having it corrected. If AI does those tasks and juniors only approve the results, they never build the judgement that approving well requires. Seniors at least learned the skill before.

This matters for the business, not just the individual. Today's seniors were yesterday's juniors. A team that skips the apprenticeship has no one ready to take over the judgement in five years.

The fix is deliberate practice. Give juniors some tasks to do by hand first, then compare their version with the AI's, and talk through the differences. That turns the AI into a tutor rather than a substitute. The changing shape of roles and approvals needs to allow for this time.

How do you use AI without losing skills?

Use AI to understand, not just to produce. Ask it why, not only what. Do some work yourself first and compare. Keep a share of core tasks done by hand. Have reviewers explain why they approved or changed something. And measure skills directly from time to time, not just output, so a decline shows up early.

1. Ask before you delegate. Use AI to explain the concept, then do the task.
2. Draft first, then compare. Write your own outline before seeing the AI's.
3. Keep some reps manual. Rotate tasks that stay hand-done, especially for juniors.
4. Explain approvals. A reviewer who must say why they approved reads more carefully.
5. Open the source. Read the original for anything that will be quoted or sent.
6. Test the skill. Occasionally run a task with AI switched off and compare.

Training helps, as long as it targets judgement. The guide to AI training for employees covers what sticks.

How can you tell if your team is deskilling?

Watch for people who cannot explain the work they approved, basic errors reaching customers, and juniors who stall without a prompt. Run an occasional task without AI and compare it with past work. Track errors found after approval. A rise in confident mistakes is often the first sign that checking skills are fading.

Perception is a poor guide here. In METR's 2025 trial, experienced developers took 19% longer with AI tools yet believed AI had made them 20% faster. People may feel sharper while getting slower or less careful, so look at the work rather than asking how people feel.

Skein offers adoption as one of three ways in, built on each team's real work, so people practise using AI to understand as well as to produce; the AI facilitation page describes the approach. That suits a team using AI daily with several junior staff. A team of senior specialists using AI lightly faces much lower risk.

Frequently asked questions

What is AI deskilling?

AI deskilling is the loss of a skill people stop practising once AI does that part of their work, such as drafting, debugging or spotting errors. Output can stay good while the AI is available and right. The risk appears when the AI is unavailable or wrong, and nobody has the skill left to notice.

Does using AI make people worse at their jobs?

It can, depending on how it is used. Early studies found doctors detecting fewer growths after AI exposure and developers learning less when they delegated code. Yet developers who used AI to ask conceptual questions scored much better. Using AI to understand, not to replace thinking, protects skills while keeping the speed.

Which jobs are most at risk of AI deskilling?

Roles where AI now does the core craft are most exposed: writing, analysis, coding, research and call summarising. Junior staff in any of these roles face the highest risk, because the tasks AI took over were how they learned. Senior staff built the skill earlier but can still lose sharpness without practice.

How do you prevent deskilling when using AI?

Ask AI to explain before you delegate, draft your own version before comparing with the AI's, and keep some tasks done by hand. Have reviewers explain their approvals, open sources before quoting them, and occasionally run tasks with AI switched off. Measuring the skill directly shows a decline before customers do.

Should junior staff use AI at work?

Yes, but with structure. Juniors should do some core tasks by hand first, then compare their work with the AI's and discuss the differences with a senior colleague. Used this way, AI acts as a tutor. Used only to produce finished work, it skips the practice juniors need to build judgement.

Related reading