{"content_id":"ucbc1pmlc9","slug":"agentic-ai-jobless-growth-white-collar-employment-risk","locale":"en","schema_type":"Report","category":"report","category_name":"Report","title":"Agentic AI and Jobless Growth: Facts and Scenarios Behind the White-Collar Job Crisis","summary":"Agentic AI may automate some office tasks and reduce new hiring, but the share of exposed tasks should not be interpreted as a layoff rate. Mass unemployment and a financial crisis are possible risk scenarios, not a predetermined future, and job quality, income, and debt data must be examined together.","author":{"name":"Injoys Editorial Team","url":"https://injoys.com/ko/about"},"key_points":["The share of jobs or tasks exposed to AI is not the same as the share of jobs that will actually disappear.","There is no established empirical evidence supporting the claim that 30–40% of white-collar jobs will become unnecessary within a few years.","The claim that U.S. mortgage balances are 12 times GDP is inconsistent with the scale shown in official statistics, and simply comparing outstanding debt with annual GDP also has limitations.","The impact of AI may first appear as reduced hiring for entry-level positions, internships, and support roles rather than immediate layoffs of existing employees.","The unemployment rate alone cannot fully capture reduced working hours, underemployment, stagnant income, and the instability of platform work."],"content_markdown":"Agentic AI is an AI system that breaks down human-defined goals into multiple steps and uses tools and data to perform work within a defined scope. This technology can reduce repetitive tasks in white-collar work, but the conclusion that it will immediately cause mass layoffs and a financial crisis has not yet been proven.\n\nThe key is to distinguish among **tasks that can technically be automated**, **tasks that companies will actually automate**, **jobs that will disappear**, and **employment changes across the economy**. Converting task exposure rates into layoff rates without these distinctions can exaggerate or misdiagnose the risks.\n\n## What Is Agentic AI?\n\nWhile conventional conversational AI focuses on answering questions or creating drafts, agentic AI is designed to carry out the following processes continuously to achieve a goal.\n\n1. Break down the goal into detailed tasks.\n2. Select the necessary information and tools.\n3. Perform actions such as searching, drafting documents, executing code, and entering information into systems.\n4. Review the results and adjust the next steps.\n5. Repeat the work until predefined termination conditions are met.\n\nCompanies can apply it to classifying customer inquiries, drafting reports, organizing sales materials, testing software, coordinating schedules, and entering data. However, because of issues involving access control, privacy protection, error propagation, auditability, and legal liability, fully autonomous operation is not always economical or safe.\n\n## Task Automation and Job Displacement Are Different\n\nA job is a bundle of multiple tasks. Even if AI handles research or document summarization, people can continue to persuade customers, grant final approval, handle exceptions, and make accountability judgments. Conversely, if automatable tasks are central to a job and a company redesigns its workflows as well, this can lead to workforce reductions.\n\n| Concept | Meaning | Correct interpretation |\n|---|---|---|\n| AI exposure | The possibility that AI may affect some tasks within a job | Does not mean layoffs are certain |\n| Automation | A system directly performs tasks previously done by people | Actual adoption requires meeting cost, accuracy, and regulatory conditions |\n| Augmentation | AI expands human productivity or the scope of work | The same workforce can produce more output |\n| Job displacement | Labor demand for the job materially declines | Accompanied by organizational restructuring and hiring or layoff decisions |\n| Net employment effect | The difference between jobs lost and jobs created | Varies by industry, occupation, region, and period |\n\nThe IMF analyzed that approximately 40% of global employment could be affected by AI, with higher exposure in advanced economies. This is an estimate of potential impact, not a prediction that the same percentage of workers will lose their jobs. In some exposed work, AI may augment labor and increase productivity and wages.\n\nThe ILO’s occupation-level analysis also finds that clerical and administrative occupations have relatively high exposure to generative AI. However, it generally assesses that changes in job content are more likely than the disappearance of entire jobs.\n\n## Major Claims That Require Verification\n\n### The Claim That 30–50% of White-Collar Work Is Meaningless Clicking\n\nThe fact that white-collar work includes many repetitive data-entry, reporting, and approval procedures is separate from the judgment that all such work is useless. Some procedures are necessary even if they do not directly produce output, including controls, record retention, and regulatory compliance.\n\nThe claim that a particular Stanford study defined 30–50% of all white-collar work as meaningless clicking is difficult to verify unless the paper’s title, study population, and measurement methods are provided. The fact that companies continued operating during periods of remote work also does not prove that the same proportion of all white-collar jobs is unnecessary.\n\n### The Claim That 30–40% of White-Collar Workers Will Be Laid Off Within 3–5 Years\n\nCurrent major studies estimate AI exposure, the potential for task automation, or companies’ hiring plans. These figures cannot be directly converted into actual layoff rates. Adoption costs, error rates, regulations, wages, customer preferences, labor contracts, and new demand all affect actual employment outcomes.\n\nThe World Economic Forum’s 2025 employer survey compiled responses indicating that structural changes could create 170 million jobs and displace 92 million jobs by 2030. This is a scenario combining the outlooks of participating companies, not a definitive prediction or an estimate of AI’s effects alone.\n\n### The Claim That Half of Graduates From Prestigious Universities Cannot Find Jobs\n\nA conclusion that half of the graduates from a particular university or degree program could not find jobs requires consistent survey criteria. Employment rates vary greatly depending on the timing of the survey, response rate, whether entrepreneurship and further education are included, and how people who have stopped looking for work are treated. Figures without a verified official employment report or study population are difficult to use as evidence for the overall labor market.\n\n## Why New Hiring May Decline First\n\nIn the early stages of AI adoption, it is easier for companies to leave vacancies unfilled or reduce entry-level hiring than to dismiss existing employees. Entry-level tasks involving substantial digital information processing—such as internships, research assistance, information gathering, basic coding, and document review—may be particularly affected.\n\nThis change poses a long-term problem. Entry-level work is both a basic production activity and a training process through which new workers learn industry knowledge and judgment. If companies automate only entry-level tasks without rebuilding their talent development systems, they may later face a shortage of skilled mid-career workers.\n\nHowever, there is no basis for generalizing that all internship and research assistant work has already been replaced by AI. The level of AI use, security restrictions, and verification costs vary greatly across organizations and fields.\n\n## How AI Investment Could Produce Jobless Growth\n\nInvestment in data centers, semiconductors, power grids, and software can increase construction and capital expenditures, contributing to GDP. If AI raises the productivity of existing workers, it can also enable more output to be produced with the same amount of labor time.\n\nHowever, GDP growth does not automatically guarantee income growth for most households. If ownership of AI infrastructure and intellectual property is concentrated among a small number of companies and investors while labor demand does not increase sufficiently, the gap between capital income and labor income may widen.\n\n| How GDP can grow while economic conditions feel weak | Indicators to monitor |\n|---|---|\n| Capital-intensive AI investment increases, but creates little permanent employment | Capital expenditures, employment by industry, post-construction operating staff |\n| Productivity gains are allocated more to profits than wages | Labor income share, real wages, corporate profits |\n| High-wage professionals benefit from AI augmentation while middle-tier jobs contract | Wages and job postings by occupation, changes by wage percentile |\n| Declining entry-level employment delays young people’s entry into the labor market | Youth employment rate, time to first job, internship and entry-level postings |\n| Regional data center investment is concentrated in specific areas | Investment by region, electricity costs, tax revenue and employment effects |\n\nA particular company’s data center plans or announced investment amounts must be distinguished from actual expenditures. When comparing total project costs with a country’s GDP, it is also necessary to determine whether the spending is spread over several years, how much imported equipment is included, and whether there is any double counting.\n\n## Changes the Unemployment Rate May Miss\n\nIn the U.S. Bureau of Labor Statistics household survey, people who worked at least one hour for pay or profit during the survey reference week are classified as employed. Under U.S. standards, unpaid family workers must work at least 15 hours in a family business to be classified as employed. Therefore, it is inaccurate to say that anyone who does even a small amount of unpaid work is considered employed.\n\nAn unemployed person is not simply someone without a job. Under U.S. standards, a person must have no job, be available for work, and have actively looked for work during the previous four weeks. Someone who gives up looking for work may move from the unemployed population to the economically inactive population.\n\nCounting gig workers as employed is not itself a statistical error. The problem is that the unemployment rate alone cannot measure the quantity and quality of work. The following indicators should also be considered.\n\n- Labor force participation rate and employment rate\n- Weekly working hours and involuntary part-time employment\n- U-6, the broad measure of labor underutilization\n- Proportions of multiple-job holders and self-employed workers\n- Real hourly wages and household income\n- Discouraged workers and the long-term unemployed\n- Coverage of social protections such as unemployment insurance, paid leave, and retirement benefits\n\n## How Are Weaker Consumption and a Debt Crisis Connected?\n\nIf AI substantially increases unemployment among high-income white-collar workers, it could affect consumption and debt repayment. However, several conditions would need to be met simultaneously for this to lead to a financial crisis.\n\n1. The employment shock is sufficiently large and persistent.\n2. Displaced workers lack sufficient liquid assets and unemployment benefits.\n3. Delinquencies spread across mortgages, auto loans, credit cards, and other debt.\n4. Falling home prices reduce available collateral.\n5. Financial institutions’ loss-absorption capacity and credit supply weaken.\n6. Fiscal and monetary policy fail to cushion the shock.\n\nThe 2008 financial crisis was not caused simply by rising unemployment. Weak lending standards, falling home prices, complex securitized products, financial institutions’ leverage, and reliance on short-term funding all contributed. Even if an AI-driven employment shock occurs in the future, it cannot be assumed that events will unfold exactly as they did in 2008.\n\n### Is the Claim That the U.S. Mortgage Market Is 12 Times GDP True?\n\nIt is inconsistent with the scale shown in official statistics. According to New York Fed data, outstanding U.S. household mortgage balances were approximately $12.6 trillion in the fourth quarter of 2024, while nominal U.S. GDP for the same year, as measured by the BEA, was approximately $29 trillion. Outstanding mortgage balances were less than half of annual GDP, not 12 times GDP.\n\nMore fundamentally, mortgages are a stock of outstanding debt at a particular point in time, while GDP is a flow of value added produced over a period. The two figures can be used for a rough comparison of the scale of risk, but they should not be interpreted as the same type of indicator. Assessing financial stability requires considering delinquency rates, borrower creditworthiness, loan-to-value ratios, home prices, and banks’ capital ratios together.\n\n## The Federal Reserve’s Ability to Respond Cannot Be Predetermined\n\nThe Federal Reserve conducts monetary policy under authority granted by law, but debates over political pressure and institutional independence may continue. A claim that a crisis will occur in a particular year and that the central bank will fail to respond appropriately is a scenario that includes political and institutional assumptions beyond an economic forecast.\n\nCrisis response also involves more than policy rates and quantitative easing. Liquidity provision, supervision of financial institutions, deposit protection, fiscal support, and unemployment safety nets all work together. Policy effectiveness varies depending on inflation, government debt, and the soundness of financial institutions at the time.\n\n## Early-Warning Indicators for Assessing Risk\n\nWhether AI is actually intensifying jobless growth should preferably be assessed through simultaneous changes in the following indicators.\n\n| Area | Early-warning indicator | Example of a risk signal |\n|---|---|---|\n| Hiring | Entry-level and internship postings, hiring rate, vacancy rate | Entry-level hiring declines persistently even as total employment holds steady |\n| Employment quality | Working hours, involuntary part-time work, U-6 | Working hours and income fall even though the number of employed people does not |\n| Distribution of productivity gains | Productivity, real wages, labor income share | Productivity rises while median wages stagnate |\n| Household finances | Credit card, auto loan, and mortgage delinquency rates | Delinquencies rise simultaneously across several types of loans |\n| Corporate behavior | AI investment, labor costs, headcount by job category | AI spending increases alongside white-collar workforce reductions |\n| Youth entry | Youth employment rate, time to first job | Post-graduation unemployment becomes structurally prolonged |\n\nThe direction of the macroeconomy should not be determined based on one or two layoff announcements or AI investment plans. Employment data by industry, corporate finances, household credit, wages, and working hours must be tracked together.\n\n## Responses That Businesses and Policymakers Can Prepare\n\n### Businesses\n\n- Evaluate the benefits of automation and the costs of errors at the task level, rather than for entire jobs.\n- Design training and apprenticeship programs to replace entry-level tasks eliminated by AI.\n- In addition to workforce reductions, consider shorter working hours, internal transfers, and sharing productivity gains.\n- Require human approval, logging, and appeal procedures for high-risk work.\n\n### Governments and Educational Institutions\n\n- Rapidly measure working hours, income, underemployment, and platform labor in addition to unemployment.\n- Link transitional income support and retraining to actual hiring demand.\n- Expand paid apprenticeships, internships, and on-the-job training so young people can begin their careers.\n- Assess the distribution of productivity gains and electricity and regional costs resulting from AI infrastructure investment.\n- Review systems so that unemployment insurance and social insurance also cover temporary, multiple-job, and platform workers.\n\n## Conclusion\n\nThe risk that agentic AI will put pressure on white-collar and youth employment while concentrating productivity gains among capital owners is a real policy challenge. In particular, declining new hiring may appear before mass layoffs while also undermining the development of the future skilled workforce.\n\nHowever, claims that 30–40% of white-collar jobs will soon disappear or that the U.S. mortgage market is 12 times GDP require verification of their evidence and units. AI exposure is not the unemployment rate, and GDP growth does not automatically determine employment, wages, or consumption outcomes. What is needed is not definitive doomsaying, but continuous monitoring that connects changes in tasks, hiring, working hours, income distribution, and delinquency data.","content_html":"\u003cp\u003eAgentic AI is an AI system that breaks down human-defined goals into multiple steps and uses tools and data to perform work within a defined scope. This technology can reduce repetitive tasks in white-collar work, but the conclusion that it will immediately cause mass layoffs and a financial crisis has not yet been proven.\u003c/p\u003e\n\u003cp\u003eThe key is to distinguish among \u003cstrong\u003etasks that can technically be automated\u003c/strong\u003e, \u003cstrong\u003etasks that companies will actually automate\u003c/strong\u003e, \u003cstrong\u003ejobs that will disappear\u003c/strong\u003e, and \u003cstrong\u003eemployment changes across the economy\u003c/strong\u003e. Converting task exposure rates into layoff rates without these distinctions can exaggerate or misdiagnose the risks.\u003c/p\u003e\n\u003ch2\u003e\n\u003ca href=\"#what-is-agentic-ai\" class=\"anchor\" id=\"what-is-agentic-ai\"\u003e\u003c/a\u003eWhat Is Agentic AI?\u003c/h2\u003e\n\u003cp\u003eWhile conventional conversational AI focuses on answering questions or creating drafts, agentic AI is designed to carry out the following processes continuously to achieve a goal.\u003c/p\u003e\n\u003col\u003e\n\u003cli\u003eBreak down the goal into detailed tasks.\u003c/li\u003e\n\u003cli\u003eSelect the necessary information and tools.\u003c/li\u003e\n\u003cli\u003ePerform actions such as searching, drafting documents, executing code, and entering information into systems.\u003c/li\u003e\n\u003cli\u003eReview the results and adjust the next steps.\u003c/li\u003e\n\u003cli\u003eRepeat the work until predefined termination conditions are met.\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eCompanies can apply it to classifying customer inquiries, drafting reports, organizing sales materials, testing software, coordinating schedules, and entering data. However, because of issues involving access control, privacy protection, error propagation, auditability, and legal liability, fully autonomous operation is not always economical or safe.\u003c/p\u003e\n\u003ch2\u003e\n\u003ca href=\"#task-automation-and-job-displacement-are-different\" class=\"anchor\" id=\"task-automation-and-job-displacement-are-different\"\u003e\u003c/a\u003eTask Automation and Job Displacement Are Different\u003c/h2\u003e\n\u003cp\u003eA job is a bundle of multiple tasks. Even if AI handles research or document summarization, people can continue to persuade customers, grant final approval, handle exceptions, and make accountability judgments. Conversely, if automatable tasks are central to a job and a company redesigns its workflows as well, this can lead to workforce reductions.\u003c/p\u003e\n\u003cdiv class=\"overflow-x-auto\"\u003e\u003ctable\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth\u003eConcept\u003c/th\u003e\n\u003cth\u003eMeaning\u003c/th\u003e\n\u003cth\u003eCorrect interpretation\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Concept\"\u003eAI exposure\u003c/td\u003e\n\u003ctd data-label=\"Meaning\"\u003eThe possibility that AI may affect some tasks within a job\u003c/td\u003e\n\u003ctd data-label=\"Correct interpretation\"\u003eDoes not mean layoffs are certain\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Concept\"\u003eAutomation\u003c/td\u003e\n\u003ctd data-label=\"Meaning\"\u003eA system directly performs tasks previously done by people\u003c/td\u003e\n\u003ctd data-label=\"Correct interpretation\"\u003eActual adoption requires meeting cost, accuracy, and regulatory conditions\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Concept\"\u003eAugmentation\u003c/td\u003e\n\u003ctd data-label=\"Meaning\"\u003eAI expands human productivity or the scope of work\u003c/td\u003e\n\u003ctd data-label=\"Correct interpretation\"\u003eThe same workforce can produce more output\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Concept\"\u003eJob displacement\u003c/td\u003e\n\u003ctd data-label=\"Meaning\"\u003eLabor demand for the job materially declines\u003c/td\u003e\n\u003ctd data-label=\"Correct interpretation\"\u003eAccompanied by organizational restructuring and hiring or layoff decisions\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Concept\"\u003eNet employment effect\u003c/td\u003e\n\u003ctd data-label=\"Meaning\"\u003eThe difference between jobs lost and jobs created\u003c/td\u003e\n\u003ctd data-label=\"Correct interpretation\"\u003eVaries by industry, occupation, region, and period\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\u003c/div\u003e\n\u003cp\u003eThe IMF analyzed that approximately 40% of global employment could be affected by AI, with higher exposure in advanced economies. This is an estimate of potential impact, not a prediction that the same percentage of workers will lose their jobs. In some exposed work, AI may augment labor and increase productivity and wages.\u003c/p\u003e\n\u003cp\u003eThe ILO’s occupation-level analysis also finds that clerical and administrative occupations have relatively high exposure to generative AI. However, it generally assesses that changes in job content are more likely than the disappearance of entire jobs.\u003c/p\u003e\n\u003ch2\u003e\n\u003ca href=\"#major-claims-that-require-verification\" class=\"anchor\" id=\"major-claims-that-require-verification\"\u003e\u003c/a\u003eMajor Claims That Require Verification\u003c/h2\u003e\n\u003ch3\u003e\n\u003ca href=\"#the-claim-that-3050-of-white-collar-work-is-meaningless-clicking\" class=\"anchor\" id=\"the-claim-that-3050-of-white-collar-work-is-meaningless-clicking\"\u003e\u003c/a\u003eThe Claim That 30–50% of White-Collar Work Is Meaningless Clicking\u003c/h3\u003e\n\u003cp\u003eThe fact that white-collar work includes many repetitive data-entry, reporting, and approval procedures is separate from the judgment that all such work is useless. Some procedures are necessary even if they do not directly produce output, including controls, record retention, and regulatory compliance.\u003c/p\u003e\n\u003cp\u003eThe claim that a particular Stanford study defined 30–50% of all white-collar work as meaningless clicking is difficult to verify unless the paper’s title, study population, and measurement methods are provided. The fact that companies continued operating during periods of remote work also does not prove that the same proportion of all white-collar jobs is unnecessary.\u003c/p\u003e\n\u003ch3\u003e\n\u003ca href=\"#the-claim-that-3040-of-white-collar-workers-will-be-laid-off-within-35-years\" class=\"anchor\" id=\"the-claim-that-3040-of-white-collar-workers-will-be-laid-off-within-35-years\"\u003e\u003c/a\u003eThe Claim That 30–40% of White-Collar Workers Will Be Laid Off Within 3–5 Years\u003c/h3\u003e\n\u003cp\u003eCurrent major studies estimate AI exposure, the potential for task automation, or companies’ hiring plans. These figures cannot be directly converted into actual layoff rates. Adoption costs, error rates, regulations, wages, customer preferences, labor contracts, and new demand all affect actual employment outcomes.\u003c/p\u003e\n\u003cp\u003eThe World Economic Forum’s 2025 employer survey compiled responses indicating that structural changes could create 170 million jobs and displace 92 million jobs by 2030. This is a scenario combining the outlooks of participating companies, not a definitive prediction or an estimate of AI’s effects alone.\u003c/p\u003e\n\u003ch3\u003e\n\u003ca href=\"#the-claim-that-half-of-graduates-from-prestigious-universities-cannot-find-jobs\" class=\"anchor\" id=\"the-claim-that-half-of-graduates-from-prestigious-universities-cannot-find-jobs\"\u003e\u003c/a\u003eThe Claim That Half of Graduates From Prestigious Universities Cannot Find Jobs\u003c/h3\u003e\n\u003cp\u003eA conclusion that half of the graduates from a particular university or degree program could not find jobs requires consistent survey criteria. Employment rates vary greatly depending on the timing of the survey, response rate, whether entrepreneurship and further education are included, and how people who have stopped looking for work are treated. Figures without a verified official employment report or study population are difficult to use as evidence for the overall labor market.\u003c/p\u003e\n\u003ch2\u003e\n\u003ca href=\"#why-new-hiring-may-decline-first\" class=\"anchor\" id=\"why-new-hiring-may-decline-first\"\u003e\u003c/a\u003eWhy New Hiring May Decline First\u003c/h2\u003e\n\u003cp\u003eIn the early stages of AI adoption, it is easier for companies to leave vacancies unfilled or reduce entry-level hiring than to dismiss existing employees. Entry-level tasks involving substantial digital information processing—such as internships, research assistance, information gathering, basic coding, and document review—may be particularly affected.\u003c/p\u003e\n\u003cp\u003eThis change poses a long-term problem. Entry-level work is both a basic production activity and a training process through which new workers learn industry knowledge and judgment. If companies automate only entry-level tasks without rebuilding their talent development systems, they may later face a shortage of skilled mid-career workers.\u003c/p\u003e\n\u003cp\u003eHowever, there is no basis for generalizing that all internship and research assistant work has already been replaced by AI. The level of AI use, security restrictions, and verification costs vary greatly across organizations and fields.\u003c/p\u003e\n\u003ch2\u003e\n\u003ca href=\"#how-ai-investment-could-produce-jobless-growth\" class=\"anchor\" id=\"how-ai-investment-could-produce-jobless-growth\"\u003e\u003c/a\u003eHow AI Investment Could Produce Jobless Growth\u003c/h2\u003e\n\u003cp\u003eInvestment in data centers, semiconductors, power grids, and software can increase construction and capital expenditures, contributing to GDP. If AI raises the productivity of existing workers, it can also enable more output to be produced with the same amount of labor time.\u003c/p\u003e\n\u003cp\u003eHowever, GDP growth does not automatically guarantee income growth for most households. If ownership of AI infrastructure and intellectual property is concentrated among a small number of companies and investors while labor demand does not increase sufficiently, the gap between capital income and labor income may widen.\u003c/p\u003e\n\u003cdiv class=\"overflow-x-auto\"\u003e\u003ctable\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth\u003eHow GDP can grow while economic conditions feel weak\u003c/th\u003e\n\u003cth\u003eIndicators to monitor\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"How GDP can grow while economic conditions feel weak\"\u003eCapital-intensive AI investment increases, but creates little permanent employment\u003c/td\u003e\n\u003ctd data-label=\"Indicators to monitor\"\u003eCapital expenditures, employment by industry, post-construction operating staff\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"How GDP can grow while economic conditions feel weak\"\u003eProductivity gains are allocated more to profits than wages\u003c/td\u003e\n\u003ctd data-label=\"Indicators to monitor\"\u003eLabor income share, real wages, corporate profits\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"How GDP can grow while economic conditions feel weak\"\u003eHigh-wage professionals benefit from AI augmentation while middle-tier jobs contract\u003c/td\u003e\n\u003ctd data-label=\"Indicators to monitor\"\u003eWages and job postings by occupation, changes by wage percentile\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"How GDP can grow while economic conditions feel weak\"\u003eDeclining entry-level employment delays young people’s entry into the labor market\u003c/td\u003e\n\u003ctd data-label=\"Indicators to monitor\"\u003eYouth employment rate, time to first job, internship and entry-level postings\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"How GDP can grow while economic conditions feel weak\"\u003eRegional data center investment is concentrated in specific areas\u003c/td\u003e\n\u003ctd data-label=\"Indicators to monitor\"\u003eInvestment by region, electricity costs, tax revenue and employment effects\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\u003c/div\u003e\n\u003cp\u003eA particular company’s data center plans or announced investment amounts must be distinguished from actual expenditures. When comparing total project costs with a country’s GDP, it is also necessary to determine whether the spending is spread over several years, how much imported equipment is included, and whether there is any double counting.\u003c/p\u003e\n\u003ch2\u003e\n\u003ca href=\"#changes-the-unemployment-rate-may-miss\" class=\"anchor\" id=\"changes-the-unemployment-rate-may-miss\"\u003e\u003c/a\u003eChanges the Unemployment Rate May Miss\u003c/h2\u003e\n\u003cp\u003eIn the U.S. Bureau of Labor Statistics household survey, people who worked at least one hour for pay or profit during the survey reference week are classified as employed. Under U.S. standards, unpaid family workers must work at least 15 hours in a family business to be classified as employed. Therefore, it is inaccurate to say that anyone who does even a small amount of unpaid work is considered employed.\u003c/p\u003e\n\u003cp\u003eAn unemployed person is not simply someone without a job. Under U.S. standards, a person must have no job, be available for work, and have actively looked for work during the previous four weeks. Someone who gives up looking for work may move from the unemployed population to the economically inactive population.\u003c/p\u003e\n\u003cp\u003eCounting gig workers as employed is not itself a statistical error. The problem is that the unemployment rate alone cannot measure the quantity and quality of work. The following indicators should also be considered.\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eLabor force participation rate and employment rate\u003c/li\u003e\n\u003cli\u003eWeekly working hours and involuntary part-time employment\u003c/li\u003e\n\u003cli\u003eU-6, the broad measure of labor underutilization\u003c/li\u003e\n\u003cli\u003eProportions of multiple-job holders and self-employed workers\u003c/li\u003e\n\u003cli\u003eReal hourly wages and household income\u003c/li\u003e\n\u003cli\u003eDiscouraged workers and the long-term unemployed\u003c/li\u003e\n\u003cli\u003eCoverage of social protections such as unemployment insurance, paid leave, and retirement benefits\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch2\u003e\n\u003ca href=\"#how-are-weaker-consumption-and-a-debt-crisis-connected\" class=\"anchor\" id=\"how-are-weaker-consumption-and-a-debt-crisis-connected\"\u003e\u003c/a\u003eHow Are Weaker Consumption and a Debt Crisis Connected?\u003c/h2\u003e\n\u003cp\u003eIf AI substantially increases unemployment among high-income white-collar workers, it could affect consumption and debt repayment. However, several conditions would need to be met simultaneously for this to lead to a financial crisis.\u003c/p\u003e\n\u003col\u003e\n\u003cli\u003eThe employment shock is sufficiently large and persistent.\u003c/li\u003e\n\u003cli\u003eDisplaced workers lack sufficient liquid assets and unemployment benefits.\u003c/li\u003e\n\u003cli\u003eDelinquencies spread across mortgages, auto loans, credit cards, and other debt.\u003c/li\u003e\n\u003cli\u003eFalling home prices reduce available collateral.\u003c/li\u003e\n\u003cli\u003eFinancial institutions’ loss-absorption capacity and credit supply weaken.\u003c/li\u003e\n\u003cli\u003eFiscal and monetary policy fail to cushion the shock.\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eThe 2008 financial crisis was not caused simply by rising unemployment. Weak lending standards, falling home prices, complex securitized products, financial institutions’ leverage, and reliance on short-term funding all contributed. Even if an AI-driven employment shock occurs in the future, it cannot be assumed that events will unfold exactly as they did in 2008.\u003c/p\u003e\n\u003ch3\u003e\n\u003ca href=\"#is-the-claim-that-the-us-mortgage-market-is-12-times-gdp-true\" class=\"anchor\" id=\"is-the-claim-that-the-us-mortgage-market-is-12-times-gdp-true\"\u003e\u003c/a\u003eIs the Claim That the U.S. Mortgage Market Is 12 Times GDP True?\u003c/h3\u003e\n\u003cp\u003eIt is inconsistent with the scale shown in official statistics. According to New York Fed data, outstanding U.S. household mortgage balances were approximately $12.6 trillion in the fourth quarter of 2024, while nominal U.S. GDP for the same year, as measured by the BEA, was approximately $29 trillion. Outstanding mortgage balances were less than half of annual GDP, not 12 times GDP.\u003c/p\u003e\n\u003cp\u003eMore fundamentally, mortgages are a stock of outstanding debt at a particular point in time, while GDP is a flow of value added produced over a period. The two figures can be used for a rough comparison of the scale of risk, but they should not be interpreted as the same type of indicator. Assessing financial stability requires considering delinquency rates, borrower creditworthiness, loan-to-value ratios, home prices, and banks’ capital ratios together.\u003c/p\u003e\n\u003ch2\u003e\n\u003ca href=\"#the-federal-reserves-ability-to-respond-cannot-be-predetermined\" class=\"anchor\" id=\"the-federal-reserves-ability-to-respond-cannot-be-predetermined\"\u003e\u003c/a\u003eThe Federal Reserve’s Ability to Respond Cannot Be Predetermined\u003c/h2\u003e\n\u003cp\u003eThe Federal Reserve conducts monetary policy under authority granted by law, but debates over political pressure and institutional independence may continue. A claim that a crisis will occur in a particular year and that the central bank will fail to respond appropriately is a scenario that includes political and institutional assumptions beyond an economic forecast.\u003c/p\u003e\n\u003cp\u003eCrisis response also involves more than policy rates and quantitative easing. Liquidity provision, supervision of financial institutions, deposit protection, fiscal support, and unemployment safety nets all work together. Policy effectiveness varies depending on inflation, government debt, and the soundness of financial institutions at the time.\u003c/p\u003e\n\u003ch2\u003e\n\u003ca href=\"#early-warning-indicators-for-assessing-risk\" class=\"anchor\" id=\"early-warning-indicators-for-assessing-risk\"\u003e\u003c/a\u003eEarly-Warning Indicators for Assessing Risk\u003c/h2\u003e\n\u003cp\u003eWhether AI is actually intensifying jobless growth should preferably be assessed through simultaneous changes in the following indicators.\u003c/p\u003e\n\u003cdiv class=\"overflow-x-auto\"\u003e\u003ctable\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth\u003eArea\u003c/th\u003e\n\u003cth\u003eEarly-warning indicator\u003c/th\u003e\n\u003cth\u003eExample of a risk signal\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Area\"\u003eHiring\u003c/td\u003e\n\u003ctd data-label=\"Early-warning indicator\"\u003eEntry-level and internship postings, hiring rate, vacancy rate\u003c/td\u003e\n\u003ctd data-label=\"Example of a risk signal\"\u003eEntry-level hiring declines persistently even as total employment holds steady\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Area\"\u003eEmployment quality\u003c/td\u003e\n\u003ctd data-label=\"Early-warning indicator\"\u003eWorking hours, involuntary part-time work, U-6\u003c/td\u003e\n\u003ctd data-label=\"Example of a risk signal\"\u003eWorking hours and income fall even though the number of employed people does not\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Area\"\u003eDistribution of productivity gains\u003c/td\u003e\n\u003ctd data-label=\"Early-warning indicator\"\u003eProductivity, real wages, labor income share\u003c/td\u003e\n\u003ctd data-label=\"Example of a risk signal\"\u003eProductivity rises while median wages stagnate\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Area\"\u003eHousehold finances\u003c/td\u003e\n\u003ctd data-label=\"Early-warning indicator\"\u003eCredit card, auto loan, and mortgage delinquency rates\u003c/td\u003e\n\u003ctd data-label=\"Example of a risk signal\"\u003eDelinquencies rise simultaneously across several types of loans\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Area\"\u003eCorporate behavior\u003c/td\u003e\n\u003ctd data-label=\"Early-warning indicator\"\u003eAI investment, labor costs, headcount by job category\u003c/td\u003e\n\u003ctd data-label=\"Example of a risk signal\"\u003eAI spending increases alongside white-collar workforce reductions\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd data-label=\"Area\"\u003eYouth entry\u003c/td\u003e\n\u003ctd data-label=\"Early-warning indicator\"\u003eYouth employment rate, time to first job\u003c/td\u003e\n\u003ctd data-label=\"Example of a risk signal\"\u003ePost-graduation unemployment becomes structurally prolonged\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\u003c/div\u003e\n\u003cp\u003eThe direction of the macroeconomy should not be determined based on one or two layoff announcements or AI investment plans. Employment data by industry, corporate finances, household credit, wages, and working hours must be tracked together.\u003c/p\u003e\n\u003ch2\u003e\n\u003ca href=\"#responses-that-businesses-and-policymakers-can-prepare\" class=\"anchor\" id=\"responses-that-businesses-and-policymakers-can-prepare\"\u003e\u003c/a\u003eResponses That Businesses and Policymakers Can Prepare\u003c/h2\u003e\n\u003ch3\u003e\n\u003ca href=\"#businesses\" class=\"anchor\" id=\"businesses\"\u003e\u003c/a\u003eBusinesses\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003eEvaluate the benefits of automation and the costs of errors at the task level, rather than for entire jobs.\u003c/li\u003e\n\u003cli\u003eDesign training and apprenticeship programs to replace entry-level tasks eliminated by AI.\u003c/li\u003e\n\u003cli\u003eIn addition to workforce reductions, consider shorter working hours, internal transfers, and sharing productivity gains.\u003c/li\u003e\n\u003cli\u003eRequire human approval, logging, and appeal procedures for high-risk work.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch3\u003e\n\u003ca href=\"#governments-and-educational-institutions\" class=\"anchor\" id=\"governments-and-educational-institutions\"\u003e\u003c/a\u003eGovernments and Educational Institutions\u003c/h3\u003e\n\u003cul\u003e\n\u003cli\u003eRapidly measure working hours, income, underemployment, and platform labor in addition to unemployment.\u003c/li\u003e\n\u003cli\u003eLink transitional income support and retraining to actual hiring demand.\u003c/li\u003e\n\u003cli\u003eExpand paid apprenticeships, internships, and on-the-job training so young people can begin their careers.\u003c/li\u003e\n\u003cli\u003eAssess the distribution of productivity gains and electricity and regional costs resulting from AI infrastructure investment.\u003c/li\u003e\n\u003cli\u003eReview systems so that unemployment insurance and social insurance also cover temporary, multiple-job, and platform workers.\u003c/li\u003e\n\u003c/ul\u003e\n\u003ch2\u003e\n\u003ca href=\"#conclusion\" class=\"anchor\" id=\"conclusion\"\u003e\u003c/a\u003eConclusion\u003c/h2\u003e\n\u003cp\u003eThe risk that agentic AI will put pressure on white-collar and youth employment while concentrating productivity gains among capital owners is a real policy challenge. In particular, declining new hiring may appear before mass layoffs while also undermining the development of the future skilled workforce.\u003c/p\u003e\n\u003cp\u003eHowever, claims that 30–40% of white-collar jobs will soon disappear or that the U.S. mortgage market is 12 times GDP require verification of their evidence and units. AI exposure is not the unemployment rate, and GDP growth does not automatically determine employment, wages, or consumption outcomes. What is needed is not definitive doomsaying, but continuous monitoring that connects changes in tasks, hiring, working hours, income distribution, and delinquency data.\u003c/p\u003e\n","tags":["Labor Market","Agentic AI","White collar employment","Jobless growth","US economy"],"faqs":[{"question":"How does agentic AI differ from conventional generative AI?","answer":"While conventional generative AI mainly answers questions or generates content, agentic AI performs multiple steps, such as breaking a goal down into tasks, using tools, and checking the results. However, execution authority, error handling, and the scope of human oversight vary by system."},{"question":"Is the percentage of jobs exposed to AI the projected layoff rate?","answer":"No. Exposure means that AI may affect some tasks within a job. AI may complement people, or automation may not occur because of cost, regulatory, or error-related issues, so the exposure rate should not be interpreted as the layoff rate."},{"question":"Is the projection that 30–40% of white-collar workers will become unnecessary within 3–5 years definitive?","answer":"There is no consensus empirical evidence that definitively supports that percentage and timeframe. Actual changes in employment depend not only on technological performance but also on wages, adoption costs, regulations, economic conditions, new demand, and companies' organizational redesign."},{"question":"Why might AI affect entry-level hiring before existing employees?","answer":"It is easier for companies to leave vacancies unfilled or reduce hiring for repetitive entry-level tasks than to lay off existing employees. If research, basic document drafting, and simple coding are automated, demand for interns and assistants may decline first, but this does not mean they have already been replaced in every industry."},{"question":"Are people classified as employed in the United States even if they work only 1 hour a week?","answer":"In the U.S. household survey, people may be classified as employed if they work at least 1 hour for pay or profit during the reference week. However, under U.S. standards, unpaid family workers must work at least 15 hours in a family business, and detailed criteria may vary depending on each country's statistical system."},{"question":"Does an increase in gig work make the unemployment rate a false statistic?","answer":"Classifying gig workers as employed is not itself an error. However, looking only at the unemployment rate may obscure short working hours, low income, and involuntary part-time work, so the employment rate, U-6, working hours, real income, and social insurance coverage rate should also be considered."},{"question":"Is the U.S. mortgage market 12 times GDP?","answer":"That is inconsistent with official statistics. In the fourth quarter of 2024, outstanding U.S. household mortgage debt was approximately $12.6 trillion, while nominal GDP that year was approximately $29 trillion. Mortgages are also an outstanding balance at a specific point in time, while GDP is an annual flow of production, so care is needed even when making a simple comparison in multiples."},{"question":"Could AI-driven unemployment cause a financial crisis like the one in 2008?","answer":"It is one possible risk pathway, but it is not an outcome that occurs automatically. A crisis could intensify only if widespread long-term unemployment, loan delinquencies, falling home prices, financial institution losses, and a credit crunch occur together."},{"question":"Can individuals' living standards worsen even if GDP grows?","answer":"Yes. Even if capital-intensive investment raises GDP and productivity, the perceived living standards of many households may not improve if gains are concentrated in capital income while employment and median wages stagnate."},{"question":"Which indicators would show AI's impact on employment first?","answer":"Entry-level and internship job postings, companies' hiring and vacancy rates, weekly working hours, involuntary part-time work, wages by occupation, and the youth employment rate should be considered together. Financial risk can be assessed by checking whether delinquency rates for credit cards, auto loans, and mortgages rise simultaneously."}],"sources":[{"url":"https://www.imf.org/en/Blogs/Articles/2024/01/14/ai-will-transform-the-global-economy-lets-make-sure-it-benefits-humanity","title":"IMF: AI Will Transform the Global Economy","type":"data_point"},{"url":"https://www.ilo.org/publications/generative-ai-and-jobs-refined-global-index-occupational-exposure","title":"ILO: Generative AI and Jobs — A Refined Global Index of Occupational Exposure","type":"source"},{"url":"https://www.bls.gov/cps/definitions.htm","title":"U.S. Bureau of Labor Statistics: Labor Force Statistics Definitions","type":"source"},{"url":"https://www.newyorkfed.org/microeconomics/hhdc","title":"Federal Reserve Bank of New York: Household Debt and Credit","type":"data_point"},{"url":"https://www.bea.gov/data/gdp/gross-domestic-product","title":"U.S. Bureau of Economic Analysis: Gross Domestic Product","type":"data_point"},{"url":"https://www.weforum.org/publications/the-future-of-jobs-report-2025/","title":"World Economic Forum: Future of Jobs Report 2025","type":"source"},{"url":"https://aiindex.stanford.edu/report/","title":"Stanford University: AI Index Report","type":"source"},{"url":"https://www.oecd.org/en/publications/oecd-employment-outlook-2023_08785bba-en.html","title":"OECD Employment Outlook 2023: Artificial Intelligence and the Labour Market","type":"source"}],"images":[{"id":318,"url":"https://injoys.com/rails/active_storage/blobs/proxy/eyJfcmFpbHMiOnsiZGF0YSI6MzU4MCwicHVyIjoiYmxvYl9pZCJ9fQ==--16d3897bc038e84945ac4f128915b4b9fbc21299/ai-6058d820.webp","is_representative":true,"generation_method":"ai_image","license":"ai_generated","mime_type":"image/webp","translations":{"ko":{"alt":"AI 네트워크가 문서·일정·파일·업무를 연결하고 직장인들이 일하거나 줄 서 있는 일러스트","caption":"에이전틱 AI의 업무 자동화와 변화하는 화이트칼라 고용 환경을 함께 보여준다.","description":null},"en":{"alt":"AI network linking documents, calendars, files and tasks as office workers work and queue","caption":"The illustration connects agentic AI automation with a changing white-collar employment landscape.","description":null},"ja":{"alt":"AIネットワークが文書や予定、ファイル、業務をつなぎ、働く人々と列に並ぶ人々を描いた図","caption":"エージェント型AIによる業務自動化と変化するホワイトカラー雇用を表している。","description":null},"es":{"alt":"Red de IA que conecta documentos, calendarios, archivos y tareas junto a oficinistas trabajando y en fila","caption":"La ilustración vincula la automatización con IA agéntica y los cambios en el empleo de oficina.","description":null},"id":{"alt":"Jaringan AI menghubungkan dokumen, kalender, berkas, dan tugas saat pekerja kantor bekerja dan mengantre","caption":"Ilustrasi ini mengaitkan otomatisasi AI agentik dengan perubahan lapangan kerja kerah putih.","description":null},"pt":{"alt":"Rede de IA conecta documentos, calendários, arquivos e tarefas enquanto profissionais trabalham e aguardam na fila","caption":"A ilustração relaciona a automação por IA agêntica às mudanças no emprego de colarinho branco.","description":null},"zh-hant":{"alt":"AI網路連結文件、行事曆、檔案與工作流程，辦公人員工作或排隊等候","caption":"插圖呈現代理式AI的工作自動化，以及白領就業環境的變化。","description":null}}},{"id":319,"url":"https://injoys.com/rails/active_storage/blobs/proxy/eyJfcmFpbHMiOnsiZGF0YSI6MzU4NiwicHVyIjoiYmxvYl9pZCJ9fQ==--9acef6a590f4ae78f983572771cd45ae78c9301d/ai-da57d97d.webp","is_representative":false,"generation_method":"ai_image","license":"ai_generated","mime_type":"image/webp","translations":{"ko":{"alt":"AI 서버와 상승 화살표, 빈 의자와 모래시계, 자산과 상반된 미래를 잇는 갈림길 삽화","caption":"AI 성장과 일자리 감소가 가계의 경제적 안정에 미칠 상반된 미래를 보여준다.","description":null},"en":{"alt":"Forked paths with AI servers, a rising arrow, empty chairs, an hourglass, assets, and contrasting futures","caption":"The illustration contrasts AI-driven growth with job loss and diverging household security.","description":null},"ja":{"alt":"AIサーバーと上向き矢印、空席、砂時計、資産、対照的な未来へ続く分岐路","caption":"AI主導の成長と雇用減少が家計の安定を左右する二つの未来を描いている。","description":null},"es":{"alt":"Caminos con servidores de IA, flecha ascendente, sillas vacías, reloj de arena, activos y futuros opuestos","caption":"La ilustración contrapone el crecimiento por IA con la pérdida de empleo y la seguridad familiar.","description":null},"id":{"alt":"Jalur bercabang dengan server AI, panah naik, kursi kosong, jam pasir, aset, dan dua masa depan","caption":"Ilustrasi ini membandingkan pertumbuhan berbasis AI dengan hilangnya pekerjaan dan keamanan rumah tangga.","description":null},"pt":{"alt":"Caminhos com servidores de IA, seta ascendente, cadeiras vazias, ampulheta, ativos e futuros opostos","caption":"A ilustração contrapõe o crescimento impulsionado por IA à perda de empregos e à segurança familiar.","description":null},"zh-hant":{"alt":"AI伺服器、上升箭頭、空椅、沙漏、資產與兩種未來組成的分岔道路","caption":"插畫對比AI驅動的成長、職位流失與家庭經濟安全的不同走向。","description":null}}}],"published_at":"2026-07-28T03:03:51+09:00","updated_at":"2026-07-28T03:03:51+09:00","license":"cc_by","translation_status":"reviewed","available_locales":["ko","en","ja","es"],"data_locales":["ko","en","ja","es","id","pt","zh-hant"],"url":"https://injoys.com/en/articles/agentic-ai-jobless-growth-white-collar-employment-risk"}