When an AI tool can draft text, summarize documents, generate code, analyze information, or create images, it can look as though the human skill behind that work is becoming unnecessary.
But that conclusion mixes together several different things.
A task can be automated without an entire skill disappearing. A skill can become less important in one part of a job while becoming more important somewhere else. And a job can change substantially without being eliminated.
AI is not simply replacing skills across the board. Current evidence suggests it is more often changing which parts of work people perform themselves and which capabilities matter most. Some narrow procedural skills may lose demand, while skills such as evaluation, data interpretation, problem-solving, and working effectively with AI can become more important.[1,2,3]
The useful question is therefore not only “Can AI do this task?”
It is also “What happens to the human skill once AI can help with part of it?”
What Does It Mean for AI to Replace a Skill?
The phrase “AI replacing skills” can be misleading because skills are rarely used in isolation.
Writing, for example, can involve:
- Generating sentences
- Organizing ideas
- Understanding an audience
- Checking facts
- Choosing what matters
- Revising tone
- Making judgments about meaning
An AI tool might perform one or several of those activities. That does not necessarily mean the entire skill of writing has been replaced.
So in this context, skill replacement is better understood as a reduction in the need for a person to perform a particular capability manually. The distinction matters. Typing every sentence may become less important in an AI-assisted workflow. Knowing whether the sentence is accurate, appropriate, useful, or misleading may become more important. The visible activity changes, but the underlying skill requirement does not always disappear.
Task Automation Is Not the Same as Skill Replacement
This is one of the most important boundaries in the discussion.
The International Labour Organization’s 2025 analysis examined exposure to generative AI at the level of occupational tasks rather than assuming that an exposed occupation would simply disappear.[1]
The ILO estimated that one in four workers globally were in occupations with some degree of generative-AI exposure. But because most occupations contain tasks that still require human input, the study concluded that job transformation is the more likely overall effect than complete redundancy.[1]
That does not mean no work will be automated. It means exposure is not the same thing as replacement. A job usually contains many different tasks.
AI may perform some of them well, partly assist with others, and contribute very little to the rest. When that happens, the composition of the work changes. So does the mix of skills that matters.
A Skill Can Become Less Important Without Disappearing
Calculators provide a simple example.
Most people no longer need to perform every calculation manually in ordinary life. The ability to calculate by hand still exists and remains important in education and particular settings, but routine demand for that method has declined. The task changed because the tool absorbed part of the procedure.
AI can create similar changes in cognitive work. A person who once produced every first draft manually may increasingly begin with generated material. Someone who previously formatted documents by hand may ask an AI tool to restructure them. Basic administrative or clerical work may also contain activities that can increasingly be automated.
OECD evidence already shows that AI exposure is associated with changes in demand for some general office, clerical, and basic digital skills.[3] So it would be unrealistic to say that AI never reduces the value of existing skills. Some forms of manual execution can genuinely become less important. The more interesting question is what replaces their importance.
Other Skills Can Become More Important at the Same Time
Automation does not only remove work. It can shift where skill is needed.
The OECD’s 2026 review of AI and workforce skills reports that AI adoption is increasing demand for higher-level skills in a number of workplaces. It highlights digital capability, data use and interpretation, management skills, problem-solving, creativity, and innovation as important parts of AI-exposed work.[2]
The OECD’s survey of small and medium-sized businesses found a similar pattern. Twice as many surveyed SMEs reported that generative AI increased their overall skill needs as reported a decrease, with data analysis and interpretation and creativity and innovation among the areas most often described as becoming more important.[3]
That is a more complicated picture than simple replacement. Some execution becomes easier. Some checking becomes more important. Some technical knowledge becomes less central. Some judgment or interpretation becomes more valuable.
The skill mix moves.
A Useful Mental Model: Navigation Apps and Map Skills
The navigation analogy from the original article still works well. Before turn-by-turn navigation became common, finding an unfamiliar destination could involve reading a map, remembering road names, and planning the route manually.
A navigation app performs much of that procedural work. That changes which skills are used. The driver may no longer need to memorize the route.
But the tool does not remove every decision around the journey. The person may still need to notice that a road is physically blocked, decide whether a suggested route feels appropriate, understand where they actually want to go, or recognize that the destination entered into the app was wrong.
Some map-reading ability becomes less necessary during an ordinary trip. Other forms of awareness and judgment remain. AI-assisted work can follow the same broad pattern.
The tool may absorb part of the procedure without absorbing everything the broader activity requires.
The Human Role Does Not Simply Move to “Oversight”
There is a tempting way to simplify this transition:
AI does the work, and the person checks it. Sometimes that is accurate. But it should not become another universal rule.
Different tasks reorganize differently.
Someone may use AI to generate a starting point and then perform most of the substantive work manually. Another person may delegate almost all of a repetitive task and only investigate exceptions. Someone else may use AI mainly to explore alternatives rather than produce a final output.
In some workflows, review becomes more important. In others, the tool may allow a person to perform work they previously lacked the time or specialist capability to attempt.
That is why deciding whether AI tools actually belong in everyday work is a different question from asking whether AI can technically perform a task. The presence of automation does not determine one fixed human role. It changes the division of work.
What Happens to Judgment?
Judgment is frequently described as the skill that remains once AI handles execution. There is some truth in that, but the evidence suggests something more specific.
A 2025 study of 319 knowledge workers found that generative AI changed how participants described their critical-thinking activity. Their attention shifted toward activities such as verifying information, integrating AI responses, and overseeing the overall task.[4]
The study also found that greater confidence in AI was associated with less reported critical-thinking effort, while greater confidence in one’s own ability was associated with more.[4]
This does not prove that using AI permanently weakens critical thinking.The study was based on workers’ reported experiences rather than a long-term test of skill loss. But it does show why the word replacement can be too simple. A cognitive skill may not disappear. Its point of use can move.
Instead of spending most of the effort producing an answer, someone may spend more effort deciding whether the generated answer deserves to be used.
Does Less Practice Mean Skills Will Weaken?
This is one of the harder questions.
If a tool repeatedly performs something a person previously practised manually, it is reasonable to ask what happens to that person’s ability over time. But there is not yet enough evidence to make one universal claim about generative AI and long-term deskilling across writing, coding, analysis, research, and other kinds of work. Different skills are learned differently. Different AI workflows remove different amounts of human involvement.
People also use AI in very different ways. Someone may accept generated work with little engagement. Someone else may use the same tool to compare approaches, identify weaknesses, or practise unfamiliar tasks.
Current research therefore supports a concern about how cognitive work is redistributed, but not the blanket claim that using AI inevitably weakens the underlying skill.[4] That distinction is especially important for beginners.
What About Beginners Learning a Skill?
The original article argued that beginners still need foundational understanding. The basic idea is useful, but it should not be turned into a universal rule that learning must always happen exactly as it did before AI.
AI can demonstrate examples, explain concepts, generate practice material, and help someone attempt work that might otherwise be inaccessible. At the same time, there is a meaningful difference between producing a result with assistance and being able to evaluate that result.
A beginner may receive competent-looking code without yet knowing why it works. They may receive polished prose without recognizing a factual weakness. They may get a plausible explanation without knowing which assumption is questionable.
This is where the uneven nature of AI capability matters. Why AI Tools Feel Smart but Still Make Simple Mistakes explains why an impressive result on one task does not guarantee reliable performance on another.
For learning, the useful question is not whether AI assistance is allowed. It is whether enough understanding remains with the learner to notice when the tool needs questioning.
Some Skills May Be Partly Substituted
It is also important not to overcorrect in the opposite direction. AI will not merely “assist” every skill indefinitely. Some capabilities can become less economically valuable when a tool performs them cheaply and reliably enough.
The OECD’s current evidence includes signs of declining demand for some routine office skills and reports from firms that AI can reduce certain skill requirements.[3]
The ILO also identifies occupations and tasks with much higher automation exposure than others.[1] So genuine substitution is possible.
The likely pattern, however, is not one clean list of human skills that survive and AI skills that disappear. It is more uneven. A narrow procedure may become highly automated while the broader occupation remains. One skill may decline while another becomes more important. And improvements in AI can move those boundaries again.
Job Replacement Is a Different Question
Questions about skills are often mixed together with questions about employment.
They are related, but they are not identical. A worker can keep the same job title while the actual skill mix changes substantially. A company can reduce the amount of labour required for one task without removing the entire role. A person can also gain responsibility for work they previously could not do because AI lowers the difficulty of one part of it.
The ILO specifically warns that exposure estimates describe the potential for tasks to be performed with generative AI, not actual future job losses.[1] Whether employment grows, shrinks, or reorganizes depends on much more than technical capability.
Adoption decisions matter.
Costs matter.
Demand matters.
Regulation matters.
Workplace design matters.
So the question “Is AI replacing jobs?” cannot be answered simply by counting the tasks an AI system can perform.
Is the Change Good or Bad?
There is no single answer.
A repetitive task becoming easier may free time for work that requires more thought. It can also remove an activity through which someone previously gained experience. AI may allow a less experienced person to complete work that once required specialist assistance. It may also make it harder to tell whether that person actually understands the result.
An organization may use AI to support skilled employees. Another may use the same technology primarily to reduce labour costs. The technology creates possibilities. How the work is redesigned determines much of the outcome.
That is why phrases such as “AI is replacing skills” or “AI is only augmenting people” are both too broad. Different skills can move in different directions at the same time.
A Better Way to Think About Skills in an AI-Assisted World
Instead of asking whether a skill has simply survived or disappeared, it helps to ask four narrower questions:
- Which parts of the activity can the tool now perform?
- Which parts still require human capability?
- Which existing skills are being practised less?
- Which new forms of judgment, interpretation, verification, or tool use are becoming more important?
Those questions reveal the actual shift. A writer may type fewer first-draft sentences but spend more time evaluating meaning. A researcher may spend less time producing an initial summary but more time checking evidence. A programmer may write less routine code manually while spending more effort understanding generated code and testing whether it behaves correctly.
The work has not stayed the same. But it has not necessarily lost its need for skill either.
Conclusion
Is AI replacing skills or changing how skills are used?
Current evidence points to both substitution and transformation, depending on the task.
Generative AI can reduce the need for some forms of manual or procedural work.[1,3] At the same time, AI-exposed workplaces continue to require—and in some cases report greater demand for—other capabilities such as data interpretation, problem-solving, creativity, management, and effective use of AI itself.[2,3]
The important distinction is that tasks, skills, and jobs are not interchangeable. Automating one task does not prove that an entire skill has disappeared. Keeping a job does not mean its required skills have stayed unchanged. And using AI does not automatically preserve or weaken human capability in the same way for every person.
The more realistic expectation is that AI changes the distribution of skill: what people practise themselves, what they delegate, what they need to verify, and what becomes valuable next.
References
- Gmyrek P. et al., International Labour Organization. Generative AI and Jobs: A Refined Global Index of Occupational Exposure. ILO Working Paper 140, 2025.
- OECD. AI and skills: What we know so far. 2026.
- OECD. Generative AI and the SME Workforce: New Survey Evidence. 2025.
- Lee H-P. et al. The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers. CHI 2025, 2025.
