AI is taking jobs: what to do and how to assess the risk

Olena and the soldering robot: not a real case, but a proof of concept

Olena is— a fictional composite character, not a real worker or a hidden "success story." She is 43 years old. For 15 years, she has been soldering the same connection at a factory. The work is repetitive, her eyes get tired, and the smoke and uncomfortable posture harm her health. But her salary supports the family budget, her colleagues have become part of her life, and there may be no other specialty nearby.

The enterprise buys a robotic line. It is important not to conflate technologies here: a modern generative model itself does not physically solder anything. Replacing such a task requires robotics, computer vision, process engineering, and often software with AI elements. For Olena, the difference in the name of the technology is small—her usual task disappears.

Different futures are possible from here. In one, Olena moves into quality control, trains during paid working hours, and maintains her salary. In another, she receives short-term compensation and goes to look for unstable work. The robot is the same, the productivity is the same, but the moral outcome is opposite.

First mistake: confusing a task with a profession

A profession consists of tasks. Some are easy to describe and automate; others require responsibility, manual adaptation, communication, knowledge of a specific production process, or working in an unpredictable environment. Therefore, the statement "AI impacts a profession" does not equal "the entire profession will disappear."

Update The International Labour Organization 2025 analyzes almost 30,000 tasks. According to its assessment, every fourth worker in the world is in a profession with some level of generative AI impact. However, the authors expect mostly transformation, rather than total job redundancy, because human involvement remains necessary in most professions.

Microsoft researchers analyzed 200,000 anonymized interactions with Bing Copilot and saw the greatest suitability of generative AI for information work: searching, writing, explaining, and analyzing. This is a measure of task applicability, not a forecast of layoffs. For physical labor, the pace will also be determined by the cost of robots, safety, reliability, and production restructuring.

What is already known about losses—and what we do not yet know

The optimistic phrase "technology always creates new professions" describes a possibility for the economy as a whole, but does not guarantee a good result for a specific individual. A new job might appear in a different city, require years of training, or be inaccessible due to health, childcare, or age. Growth statistics do not automatically transition Olena to a new position.

The IMF estimates, that AI will affect approximately 40% of global employment: it will complement and make part of the work more productive, while in other cases, it may reduce demand for people, wages, or the number of positions. This is an indicator of impact, not a claim that 40% of people will lose their jobs.

World Economic Forum based on a survey of over a thousand large employers, expects 170 million new roles and the displacement of 92 million by 2030 — a net gain of 78 million. But this is not a precise prediction and not a guarantee for a laid-off person: new and disappearing jobs differ by country, profession, pay, and skill requirements.

A historical study by Daron Acemoglu and Pascual Restrepo on industrial robots in the US revealed local negative effects: one additional robot per thousand workers was associated with a decrease in the employment share by approximately 0.18–0.34 percentage points and wages by 0.25–0.5%. This is an assessment of past experience in American labor markets, not a universal law and not an accurate forecast for Ukraine. But it shows: transitional losses can be real and geographically concentrated.

Thus, two extremes are equally weak. There is no evidence that "all professions will disappear tomorrow." Likewise, it is unfair to say that everyone laid off will painlessly become a robot operator. We know the direction of some changes, but the speed, new professions, and income distribution depend on the decisions of companies, workers, and governments.

The main economic question: who owns the productivity

Automation can increase output, reduce defects and costs, and eliminate hazardous operations. This is a real social benefit. But productivity is not a distribution mechanism. The economy can produce more, while the worker who helped create the process and accumulated production knowledge for years may receive less.

New technologies also change bargaining power. If an employer can quickly replace part of the tasks, it becomes harder for the worker to demand a share of the gains. Therefore, the answer is not simply to "learn for life." Skills are important, but rules regarding notice, severance, paid retraining, worker representation, and profit sharing are equally decisive.

What tech leaders say — and why they aren't prophecies

In August 2026, Bill Gates wrote that the upcoming transition could particularly hit entry- and mid-level positions, and reminded: work provides not only income, but also dignity and social connection. Among his proposals is status Human Reserved for roles where human presence is fundamental, as well as changes to the taxation of labor and capital, including possible taxes on robots or AI tokens. This is the position of a technological leader, rather than scientific consensus or a ready-made tax model.

In its own labor market preparation plan OpenAI directly acknowledges three simultaneous processes: new jobs will emerge, some existing ones will change, and others will disappear. The company proposes broader AI skills training and cooperation between the state, businesses, and unions. This is an important acknowledgment of risk from the technology developer itself, but training does not answer the question of who compensates for the loss of income during the transition.

A survey of Claude users, published in the Anthropic Economic Index in June 2026, showed a different side of expectations: over half of respondents hoped for collaboration with AI and more meaningful work, slightly more than half for the automation of tedious tasks, and about a third for a broad distribution of benefits. But these are Claude users, not a representative sample of the population.

Such statements are useful for understanding goals and fears, but companies and their executives have their own interests in technological development. A more reliable assessment arises where their forecasts are compared with independent data, transparent methodology, and real results for workers.

Four tests for a fair assessment of automation

1. Consequences. Have injuries, burnout, and shortages decreased overall? We must count not only profit and product price, but also unemployment, family health, and community costs.
2. Dignity and voice. Were changes explained in advance? Could a worker influence the new process, choose training, or decline with dignity? A person is not an extra piece of equipment.
3. Distribution of benefits. Who owns the robot and reaps the savings? If one side takes all the gains and the other takes all the risk, formal efficiency does not make the outcome fair.
4. Real freedom. The suggestion to "learn programming" is worthless without time, money, accessible education, and a real job vacancy. The alternative must be practically achievable for that specific person.

These tests do not require stopping technology. They change the unit of assessment: not "is the robot good or bad," but what social contract accompanies its emergence. This continues the questions from our analysis of free will and responsibility: if conditions heavily shape a person's possible actions, justice must take these conditions into account.

What a fair transition looks like in practice

The word "retraining" is often used as a magic answer. A real program must have a specific budget, timeframe, and next position. For the factory in our example, a minimum contract could be as follows:

• Inform and involve people in the decision to purchase the system, rather than on the day of dismissal.
• First, automate the most harmful tasks and look for labor complements: quality control, setup, maintenance, documentation.
• Provide transition income for 6–12 months, so that choices are not made under the threat of immediate poverty.
• Train during paid hours and link the course to a guaranteed interview or a specific vacancy.
• Share productivity through bonuses, profit sharing, higher pay for the new role, or a shorter work week.
• Evaluate the result after one year: how many people maintained their income, obtained stable work, and did not suffer health declines.

For small businesses, such a transition can be expensive, so part of the costs can be covered by industry funds or the state. However, a robotics subsidy without conditions regarding people effectively helps to privatize the benefits and shift the costs to society.

Will Universal Basic Income solve the problem

UBI — a regular, unconditional payment to everyone — can create a financial floor, strengthen bargaining power, and provide time for learning, care, or searching for a better job. But it does not automatically restore status, social circles, professional identity, or control over technology. The size, funding, taxes, housing prices, and the fate of other social programs matter.

One of the largest recent American experiments by OpenResearch gave thousands of low-income people $1,000 per month for three years, while the control group received $50. In employment research the main group worked an average of 1.3 hours less per week and was approximately 2 percentage points less employed. At the same time, the money increased financial stability and freedom of choice; some people used it for care, learning, or job searching.

What employees, companies, and the state can do right now

For employees: break down your profession into 10–15 specific tasks; identify repetitive, digital, and easily measurable ones; separately — those that require responsibility, local context, physical adaptation, or human trust. It is better to learn a tool through a real process rather than abstractly. Our beginner's roadmap to Vibe Coding shows how to turn an idea into a small, verified project.

For managers: Before automation, disclose not only the savings but also the plan for retraining, training, and distribution of gains. Measure quality and safety after implementation; do not transfer model or machine errors to a worker who does not control the system.

To the state and the community: calculate the impact by region and profession, make assistance rapid and portable, fund training alongside employers, and link benefits to income maintenance and employment. The International Labour Organization specifically emphasizes social dialogue—the participation of workers, employers, and the state in managing the transition.

Quick answers

Will AI take all the jobs?

There is no reliable basis to claim this. Studies more often show the automation of specific tasks and the restructuring of professions. At the same time, a portion of workers, businesses, and regions may suffer real losses—they cannot be hidden behind an average figure.

Shouldn't boring work be given to robots?

Yes, especially a dangerous and exhausting one. But a good result requires that the worker receives a real alternative, income during the transition, and a share of the benefits from increased productivity.

Will basic income save us?

It can provide a financial floor and freedom of choice, but it does not replace the meaning of work, social ties, education, and the fair distribution of property. Existing experiments are not yet a definitive test of nationwide UBI in the age of AI.

What should a worker do right now?

Break down a profession into tasks, identify vulnerable and resilient parts, study tools within workflows, and demand a specific transition plan. Personal adaptation is important, but it does not negate the responsibility of the employer and the state.

Conclusion: we should judge not the work, but the transition

If a machine takes a harmful operation away from a person, and society returns to them time, safety, stable income, and new opportunities—that is progress. If a machine takes both the operation and the means of living, and the benefit is concentrated in the owner, technological success coexists with social defeat.

The future of work is not determined by a single algorithm. It is shaped by property rules, taxes, bargaining power, product design, and the daily decisions of companies. Automation answers the question "what a machine can do." Humans must answer the harder one: what do we owe each other, now that a machine can?

Read about the psychological side of changes in a separate scientific review of AI fear, burnout, and loss of motivation.