What the data says
Generative AI is a neural network that writes text, answers questions, and creates images and code on a person’s request. Its effect on work is what researchers study most now.
The main figures come from ILO Working Paper 140, published in 2025. The authors analyzed the tasks of each occupation and estimated what share of them generative AI can perform.
The ILO’s result: AI affects roughly one in four workers worldwide, about 24%. In the group with the highest effect, where AI can take over most tasks, are 3.3% of workers.
An effect does not mean a layoff. It means that some tasks can be handed to AI or done faster with its help.
The International AI Safety Report, published on 3 February 2026, checks forecasts against facts. Its conclusion: AI has not yet affected overall employment, but in some occupations demand for beginners is falling.
Employers expect work to be restructured, not to disappear. According to the World Economic Forum’s Future of Jobs Report 2025, 170 million jobs will appear by 2030 and 92 million will disappear. This is a survey of employers, that is, their expectations, not a count.
Where the effect of AI is stronger and where weaker
AI affects office and administrative work the most. According to the ILO, the group with the highest effect most often includes data entry clerks, typists, accounting and office clerks, and secretaries.
These jobs have many template tasks with text and numbers: move data from a document to a spreadsheet, write a standard letter, reconcile invoices. A neural network does such tasks quickly.
Compared with the 2023 estimate, the effect has grown for jobs where almost all work happens on a computer. The ILO names web developers, database specialists, financial analysts and programmers.
AI affects manual work the least. The more physical tasks a job has, the lower the effect: in skilled trades, construction, machine operation and agriculture it is much smaller.
Countries differ too. In high-income countries the effect covers 34% of workers, in low-income countries 11%. Women have a higher share in the group with the highest effect, 4.7% against 2.4% for men, because women more often hold office positions.
Why AI changes jobs more often than it replaces them
A job consists of dozens of tasks. AI can take some of them, but almost every job keeps tasks that need a person: making a decision, reaching an agreement, answering for the result.
So the ILO’s main conclusion is this: AI will most likely change work, not abolish it.
Example: an accountant. AI will post a bank statement and prepare a draft report, but a person must check the figures, answer to the tax authority and explain to the manager what is happening with the money.
Example: an account manager. A neural network drafts an answer to a standard question, while the manager decides on deadlines, discounts and disputed cases.
There is one more reason. The International AI Safety Report notes that AI systems sometimes invent facts and give wrong advice, so someone who knows the field checks their output.
Beginners
So far the changes are most visible for people who are just starting. The International AI Safety Report 2026 cites studies: in occupations where AI applies, demand for beginners is falling, for example in writing work.
At the same time, employment of experienced workers in the same occupations has stayed the same or grown since the end of 2022. According to one British study, companies that use AI heavily hire for junior positions more slowly.
One explanation is simple: a beginner’s first tasks are often standard ones, and AI takes those first. It does not replace experience, knowledge of the field or responsibility for a decision.
A student or a beginner should learn to work with AI from day one and show employers that they can check its results.
Bad: “I won’t use a neural network, so I don’t forget how to think.”
Good: “The neural network makes the draft, and I check the facts, edit the text and answer for the result.”
What a worker should do
Start with your own list of tasks. Write down what you do in a week and mark the tasks with text, spreadsheets and templates: AI will speed these up first.
Then try to give one such task to a neural network and compare the result with your own.
Bad: “Help with the report.”
Good: “I am an accountant. Here is a list of payments for the month without client names or account numbers. Group the payments by category and show in a table which ones could not be assigned.”
Learn three things: setting a task, checking an answer and knowing what must not go into a neural network. These are the basics, and a free AI course with a test and a certificate covers them.
If you want to go deeper, the list of free AI courses collects university courses and marks what is free and what costs money.
Your main advantage next to AI is experience in your field. A neural network writes with confidence, but someone who knows the field will notice the error in a contract or in the figures.
What a manager should do
Look at tasks, not at positions. The ILO measures the effect of AI by tasks, and the same approach works in a company: which tasks of a department are standard and repeat every day.
Bad: “We’ll cut the department, AI will do everything.”
Good: “We’ll give AI the sorting of incoming requests and send the managers’ freed time to complex clients.”
Do not stop hiring beginners completely. If AI takes the simple tasks, junior staff need another growth path: checking AI results, working with clients, learning from experienced colleagues.
Standard work can be handed to AI agents. An AI agent is a program built on a neural network that performs a task step by step by itself: reads a request, fills in a table, prepares an answer. How a team of such agents works next to people is shown on the AI department page.