AI Has Not Taken Professions. It Is Taking the Entry Level of Careers

Companies rarely replace an entire profession with one model. They act more quietly: they stop opening vacancies, do not replace people who leave, phase out contractors and demand from newcomers skills that used to be expected from experienced workers.

In the old economy, a simple task was the ticket to entry into a profession. Write SEO text. Translate the standard description. Collect a table. Prepare a draft agreement. Check the facts. Reply to the client according to the script. Make some basic code. It was routine for the company. For a beginner, this is the entry level of a career.

In the new economy, this entry level was the first to come under pressure.

Artificial intelligence has not yet destroyed the labor market. The data doesn’t support the simple theory that millions of office workers have already been replaced en masse by chatbots. But an investigation of public cases, labor market studies and corporate statements shows a more subtle and dangerous picture: AI is not cutting out entire professions, but those tasks in which people previously learned to become specialists.

How we checked it

For this article, we separated four different phenomena that are often lumped together in public discourse.

The first is exposure, that is, the potential exposure of a profession to AI. This is not a dismissal, but an assessment of what part of the tasks can be accelerated or partially automated. A study by OpenAI and the University of Pennsylvania estimated that about 80% of the US workforce may have at least 10% of tasks affected by LLM, and about 19% of workers may have at least 50% of tasks affected; the authors did not explicitly call this a prediction of job disappearance. The IMF similarly estimated that about 40% of global employment and about 60% of employment in advanced economies is exposed to AI, but pointed out that some workers could benefit from increased productivity while others could face declines in demand, wages or hiring.

The second is direct layoffs. This is when a company lays off employees and directly links the solution to AI. There are more and more cases like this, but even here it is important to distinguish “AI named as the cause” from evidence that each specific job has been technically replaced by a model. Challenger, Gray & Christmas, in a May 2026 report, found that AI was cited as the cause of 38,579 announced cuts in the US for the month, or 40% of all cuts in May, but this is a database of announced reasons, not a forensic analysis of every job opening.

Third – hiring freeze and non-replacement. This is the quietest mechanism. The person leaves, but the company does not re-open the position. Headcount is reduced not through dramatic dismissal, but through natural attrition, automation and a ban on a new headcount.

Fourth – budget redistribution. The company is eliminating some roles and at the same time hiring others: engineers, AI product managers, data specialists, security, compliance, sales or customer success. In statistics, this may look like “the market has not collapsed,” but for a specific junior specialist, entering the profession becomes different.

Verification map: what is actually confirmed

Klarna

Public narrative: “AI replaced 700 support staff.”

What’s confirmed: Klarna said the AI assistant handled 2.3 million conversations in the first month, took over two-thirds of support chats, and handled the equivalent of 700 full-time agents. Reuters later wrote that the workforce had been reduced from about 5,000 to 3,800, mostly through attrition rather than layoffs.

What you shouldn’t say: You can’t write that “Klarna fired 700 people because of a bot.” The “equivalent load” was confirmed, and not the direct dismissal of 700 specific workers.

Intuit

Public narrative: “1,800 people fired because of AI.”

What’s confirmed: Intuit announced that about 1,800 employees, or 10% of the workforce, will leave the company; At the same time, the company said it plans to hire approximately 1,800 new people in areas including engineering, product, sales, customer success and marketing.

What not to say: This is not a pure “AI destroyed 1,800 jobs” case. This is a redistribution of personnel and budget.

Duolingo

Public narrative: “AI has replaced translators.”

What is confirmed: Bloomberg Law reported that Duolingo offboarded about 10% of its contractors as the company expanded its use of generative AI for content creation.

What you shouldn’t say: You can’t say that the translation profession has disappeared or that companies have fired their full-time translators en masse. Confirmed impact on contractors and routine content layer.

IBM

Public narrative: “AI will replace the back office.”

What is confirmed: Reuters quoted IBM CEO Arvind Krishna as saying there would be a pause or slowdown in hiring in back-office functions and the potential replacement of about 7,800 positions with AI and automation over several years.

What is not worth saying: this is not a one-time dismissal of 7,800 people. This is an example of how AI operates through hiring freezes and non-replacement.

Amazon

Public narrative: “AI will reduce corporate workforce.”

What is confirmed: Andy Jassy wrote to employees in June 2025 that with the introduction of generative AI and agents, the company will need fewer people for some current tasks and more people for other types of work; In the coming years, Amazon expects a reduction in the total corporate workforce due to efficiency gains.

What’s not to say: This does not mean that the entire corporate workforce is being replaced by AI. This is a statement about the restructuring of work.

Shopify

Public narrative: “First AI, then new headcount.”

What’s confirmed: TechCrunch reported on Shopify CEO Tobi Lütke’s memo: Teams must show why a task can’t be completed with AI before asking for more headcount or resources.

What should not be stated: this is not a layoff case. This is a more important management signal: AI is becoming a filter on the very fact of hiring.

Microsoft

Public narrative: “Retrenchments for AI.”

What is confirmed: AP reported that Microsoft has begun laying off about 6,000 workers, nearly 3% of its workforce, amid heavy AI spending; In Washington, many software engineering and product management roles were cut.

What’s not to say: These cuts shouldn’t automatically be considered a direct replacement for AI engineers. It is rather a mixture of optimization, simplification of management layers and redistribution of capital into AI infrastructure.

The main conclusion from the table: there are strong cases, but they almost never look like a simple “bot came – human left” scene. The real mechanism is more complex: automation of routine, hiring freeze, reduction of contractors, reduction of vendor spend, budget redistribution and pressure on managers, who now have to prove the need for a new position.

Klarna: the myth of “700 laid off” and the reality of non-replacement

Klarna has become the perfect symbol of the era. In February 2024, the company said its AI assistant processed 2.3 million customer conversations in the first month, took over two-thirds of support chats, reduced the average ticket resolution time from 11 minutes to less than two, and performed the equivalent of 700 full-time agents. For investors, this sounded like a new margin formula: fewer people, faster service, lower cost-to-serve.

But this particular case shows how easily a real number turns into a bad conclusion. “The work equivalent of 700 agents” is not “700 people were fired because of AI.” Reuters later wrote that Klarna had reduced active positions from about 5,000 to 3,800, with CEO Sebastian Siemiatkowski explaining that almost all of the reduction was achieved through attrition, not layoffs: the company simply did not hire new people to replace those who left. In 2025, the story took another turn. Reuters reported that Klarna’s CEO admitted that the company may have been “over index” on using AI to cut costs and is now shifting its focus to improving products and growth; Klarna also had open positions on the jobs portal again. This doesn’t negate the effect of AI, but it does make a more mature point: support automation can dramatically reduce the need for people on standard calls, but it doesn’t replace the entire loop of trust, exceptions, and complex customer experiences.

Intuit, Duolingo, IBM: three different ways to clean the first floor

Intuit is important as a counterexample to the simplistic article. In July 2024, the company announced the exit of approximately 1,800 employees, about 10% of the workforce. But in the same document, the company said it was not making layoffs to cut costs and planned to hire about the same number of new people in other areas. This is not an “AI took away 1,800 jobs” story; this is the story of “old roles became less of a priority, new roles became more expensive and more important.”

Duolingo takes a more direct hit at the routine layer. Here we are not talking about the mass disappearance of translators, but about the reduction of contractors: Bloomberg Law wrote that about 10% of contractors were offboarded because the company no longer needs as many people for some tasks, some of which may be related to AI. This is a typical first blow: not to protected senior experts, but to external performers and repeatable work. IBM shows the quietest mechanism: not dismissal, but the absence of a new vacancy. Reuters in 2023 reported a statement from the CEO of IBM about a freeze or slowdown in hiring in back-office functions, where about 30% of non-customer-facing roles could be replaced by AI and automation within five years. For the labor market, this may be even more important than high-profile layoffs: the position does not disappear in the news, it simply does not appear.

The macro market has not collapsed yet

If you look at the entire labor market, there is no evidence of AI unemployment yet. The Federal Reserve in March 2026, using data from Lightcast and the Census Bureau, wrote that research on the current state of AI adoption and the impact on employment is at an early stage, and long-term conclusions are difficult to draw; their analysis did not show a simple picture where firms with high AI adoption sharply reduce job postings.

The New York Fed also warned in May 2026: AI exposure to a profession does not automatically mean a drop in hiring or an increase in layoffs. In the Second District, more firms reported retraining workers into AI-exposed occupations than cutting hiring, according to the New York Fed. Even the BLS’s official projections for technology roles don’t sound like the end of the professions. In 2024–2034, the BLS predicts employment growth for data scientists by 34%, and employment of software developers, quality assurance analysts and testers by 15%, which is above the economic average. But this does not mean that everything is calm. Indeed Hiring Lab wrote in January 2026 that overall hiring remains weak, vacancies in many professions are stagnant or declining, but postings with AI-related terms are growing: AI Tracker reached 4.2% of all vacancies in December 2025, and almost 45% of data & analytics postings already contained AI-related terms. That is, the market is not disappearing, it is being reshaped in favor of people who already know how to work in the new combination of “domain + data + AI”.

The problem is not in professions. The problem is entering the profession

The most alarming part is not visible in total employment, but in the age and career structure. Stanford Digital Economy Lab’s study “Canaries in the Coal Mine?” used high-frequency administrative payroll data from the largest payroll provider in the USA and studied occupations exposed to generative AI. In the November 2025 version of the study, the authors noted that workers 22–25 years old in the most AI-exposed occupations had a decline in employment since the end of 2022, while older groups in the same professions showed growth.

The latest PwC 2026 Global AI Jobs Barometer, published on June 15, 2026, reinforces this line. PwC analyzed more than 1 billion job ads across six continents and separately 2.4 million entry-level jobs in the USA. Conclusion: AI-exposed entry-level roles now require traditionally senior-level human-intensive skills – judgment, leadership, creativity or face-to-face interactions – seven times more often; At the same time, “seniorised” entry-level roles have grown by 35% since 2019, and other entry-level roles have decreased by 10%.

This is the key to the article. AI is not necessarily destroying the profession from above. He can cut it from below.

Previously, the junior analyst learned by collecting tables. Junior lawyer – on draft contracts. Junior journalist – fact-checking and retelling documents. Junior developer – on simple tasks, tests and support. Now these tasks are the first candidates for automation. Senior with AI becomes faster. Junior without experience becomes a less obvious investment.

The Productivity Paradox: AI Helps Newbies—So They May Be Less Hired

One of the most troubling effects of generative AI is that it both increases the productivity of weak workers and reduces the need for large numbers of weak workers. The NBER “Generative AI at Work” study of 5,179 customer support agents found that access to an AI tool increased productivity, measured by issues resolved per hour, by 14% on average and by 34% for novice and low-skilled workers, with minimal effect for experienced and highly skilled workers.

At the level of the individual employee, this is good news: the newcomer reaches acceptable quality faster. At the company level, this is ambiguous: if one junior with AI does a job that previously required two, the business may not fire anyone today, but not open a second vacancy tomorrow.

Freelancing shows this effect faster because there are fewer corporate buffers. A study of the online platform following the launch of ChatGPT found that freelancers in the most affected occupations experienced a decline in both employment and earnings. This is logical: if the task is standardized and purchased as a commodity, the customer transfers it to the model faster.

AI layoffs: reality, but also a language for investors

You can’t pretend that “AI layoffs” are just a media panic. Challenger, Gray & Christmas, in a report for May 2026, directly writes that AI has become the leading reason companies give for cutting jobs, and the technology sector announced 38,242 layoffs in May – the maximum for the sector since August 2024. In May, AI was cited as the cause of 38,579 cuts, 40% of all announced cuts for the month; since the beginning of 2026, AI cited in 87,714 cuts, already more than 54,836 for the whole of 2025.

But here, too, caution is needed. Challenger measures announced cuts and reasons given. This is an important signal of corporate behavior, but not proof that each of these positions has been technically replaced by a specific model. Sometimes AI is real automation. Sometimes it’s an explanation to cut costs. Sometimes it’s a way to show investors that the company is “moving towards efficiency.” Sometimes – all at once.

Microsoft shows this gray area well. AP wrote that the company was laying off about 6,000 workers, almost 3% of its workforce, amid heavy spending on AI; Both software engineering and product management roles were cut. But a straightforward conclusion, “AI has replaced these engineers,” would be too strong. A more precise formula: the company invests in AI, simplifies management layers, redistributes costs and changes the structure of teams.

New roles: not “prompt engineer”, but a layer of responsibility

The worst mistake in forecasts is trying to guess the beautiful names of future professions. “Prompt engineer” sounded like a symbol of a new era, but in practice, prompt becomes not a profession, but basic literacy. Like searching in Google, Excel or knowing how to work with tables.

Another layer is growing: AI governance, model risk, AI compliance, data quality, security, evaluation, workflow automation, human-in-the-loop operations, domain validation, AI product management. Indeed recorded an increase in vacancies with AI terms against the backdrop of weak overall hiring, and WEF in the Future of Jobs Report 2025 named AI and big data the fastest growing skill group, followed by networks and cybersecurity and technology literacy.

Regulation perpetuates this layer of responsibility. The EU AI Act classifies AI systems for employment, worker management and access to self-employment as high-risk use cases, including recruitment, selection, filtering applications, evaluation of candidates, promotion, termination, task allocation and monitoring. For high-risk AI systems, Article 14 requires human oversight, that is, the system must be designed so that people can prevent or minimize risks to health, safety and fundamental rights. In law, a similar logic has already been formulated by professional ethics. ABA Formal Opinion 512 states that lawyers using generative AI must consider the responsibilities of competency, protection of client information, client communication, and reasonable billing. In other words, a lawyer can use AI, but cannot shift professional responsibility to it. In the creative field, a similar barrier passes through authorship. U.S. The Copyright Office in 2025 indicated that outputs of generative AI can only be protected by copyright where the human author has determined sufficient expressive elements; Mere provision of prompts is insufficient. This doesn’t protect all designers and copywriters from automation, but it increases the value of the individual as writer, editor, art director and owner of the final solution.

Who looks more resilient

The most sustainable professions are not those where there is “no AI.” There will be fewer and fewer of them. More stable roles where there is a physical world, responsibility, trust, negotiation, complex context and consequences of error.

These are medicine, nursing, education with a strong human component, facility engineering, construction, energy, security, law, finance, compliance, investigative journalism, product ownership, enterprise sales, operations, procurement, system architecture, data governance and management of complex decision chains.

But “more sustainable” does not mean “no change.” The doctor will work with AI triage and tips. Lawyer – with AI drafts and checking sources. Developer – with coding agents, tests, architecture and security review. The editor is with a stream of machine text that needs to be checked, selected and turned into a responsible product. The seller is equipped with AI-analytics of the client and automated preparation of materials.

The main line is not “human versus AI”. The main line is the person who is responsible for the result versus the person who simply performs a repeatable operation.

What is now becoming a new entry into the profession

The “what to learn” block in such an article should not turn into a career guide, but without it the conclusion will be incomplete. If the first floor of the old career ladder disappears, you need to understand what becomes the new first floor.

The first is the domain. Superficial work is getting cheaper. A person who understands law, finance, medicine, logistics, fashion, industry, market, product or customer becomes more valuable because he is the one who is able to distinguish a plausible machine answer from a workable solution.

The second is verification. Source verification, working with primary documents, filings, court decisions, regulatory texts, logs, tables, APIs and statistics are becoming not a “journalistic skill”, but basic professional protection.

Third is automation. SQL, Python at the application level, APIs, no-code/low-code, tables, parsing, vector search, agent-based workflow and understanding model limitations are becoming the new office minimum for analysts, marketers, editors, product managers, operations managers and consultants.

Fourth is responsibility. In a world where generation is almost free, what becomes scarce is the person who can say, “I checked, this can be used.”

Fifth – briefcase. It’s becoming increasingly difficult for a beginner to come with the phrase “I’m ready to learn.” Companies want to see proof: project, research, code, automation, analysis, demo, public work, case. This is unfair to beginners, but it is becoming the new reality of the market.

The main risk: not unemployment, but the subsidence of the professional pyramid

The most dangerous mistake is to argue in old categories: “AI will destroy professions” or “nothing will happen.” Both formulas are too rough.

AI may not destroy the legal profession, but it may eliminate some of the tasks of junior lawyers. It may not destroy journalism, but it will make it cheaper to retell press releases. It may not destroy development, but it will reduce the number of simple tasks on which junior learned to think like an engineer. It may not destroy customer support, but leave people with only complex, conflictual and emotionally difficult requests.

In the short term, this benefits companies. Less routine. Less headcount. Faster processes. Higher revenue per employee. In the long term, a question arises that business has not yet answered: who will train future senior specialists if it is economically cheaper to outsource the initial tasks to a model?

If a junior analyst doesn’t collect data, he doesn’t learn to see errors in the data. If a junior lawyer doesn’t write drafts, he doesn’t learn how to build an argument. If a junior developer doesn’t do simple tasks, he doesn’t learn architectural thinking. If an aspiring journalist does not fact-check, he does not become an investigator.

AI has not taken over the labor market entirely. He started from the first floor. And if the first floor disappears, the entire building of professions will one day begin to sag.