Will AI Create Job Losses and Job Shortages?

Key Takeaways

• AI can reduce hiring in some jobs and raise demand for scarce skills at the same time.
• Task redesign matters more than simple claims that whole occupations will vanish.
• Training pipelines decide whether productivity gains become broader wage gains.

AI Job Losses and Job Shortages Are the Same Labor Shock

On June 17, 2026, Jeff Bezos used a VivaTech appearance in Paris to argue that artificial intelligence would create labor shortages rather than make people redundant. AI job losses and job shortages now sit beside each other in the same public debate because both can happen inside the same economy, inside the same sector, and sometimes inside the same company.

The apparent contradiction comes from treating “jobs” as fixed containers. A job is better understood as a bundle of tasks, relationships, judgment calls, compliance duties, customer expectations, software systems, and institutional memory. Artificial intelligence can remove some tasks from that bundle, speed up others, raise the quality bar for still others, and create new work around oversight, data, integration, security, customer trust, and accountability. The result may be fewer entry-level openings in one department and more vacancies for experienced people in another.

The labor market was not sending one clean signal by June 2026. The U.S. jobs report for May 2026 showed total nonfarm payroll employment rising by 172,000, with unemployment unchanged at 4.3%. That broad number did not prove that AI had no displacement effect. It showed that overall employment can grow even when certain white-collar functions shrink, financial activities decline, and companies shift hiring toward narrower skill sets.

The loss side of the debate is strongest where AI can take over repeatable information work. Customer support scripts, routine coding, document summarization, claims triage, basic translation, standard marketing copy, quality checks, form processing, and compliance preparation are all exposed because they involve text, pattern recognition, retrieval, and structured judgment. A worker does not need to lose every task for the employer to reduce hiring. If a team of 10 can complete the same workload with seven people and better software, management may slow replacement hiring long before formal layoffs appear.

The shortage side is strongest where AI raises the ceiling on what organizations try to do. Cheaper analysis can generate more projects. Faster simulation can expand product testing. Automated drafting can create more demand for review, approval, and governance. AI-assisted software tools can help small teams build services that once required larger engineering groups. These gains generate new bottlenecks: people who understand customers, domain rules, data quality, workflow design, safety controls, procurement, privacy, cybersecurity, and change management.

This table organizes the two perspectives without treating them as mutually exclusive.

The most accurate labor-market reading is not “AI will destroy work” or “AI will create more work.” The better reading is that AI changes the shape of labor demand. Where it substitutes for tasks faster than organizations create new work, job losses appear. Where it expands what firms can attempt faster than they can hire qualified people, shortages appear.

Automation Pressure Falls Heaviest on Tasks Before Occupations

The IMF analysis published in 2024 estimated that almost 40% of global employment was exposed to AI, with exposure rising to about 60% in advanced economies. The same analysis separated exposure from outcome. Some exposed jobs can gain productivity. Others face lower labor demand, reduced wages, weaker hiring, or disappearance in the most extreme cases.

That distinction matters because exposure is not destiny. A tax preparer, paralegal, radiologist, software developer, claims analyst, recruiter, journalist, translator, policy analyst, accountant, tutor, or marketing coordinator may use AI for part of the work without seeing the entire occupation vanish. The risk grows when a large share of daily tasks can be done by software, the output can be checked cheaply, errors have manageable consequences, and customers do not insist on human service.

The ILO refined index published in May 2025 treated occupational exposure as a task-level issue. That approach fits how employers actually implement AI. They rarely abolish a job title on day one. They test a chatbot in support, add document drafting to a legal team, deploy coding assistants in software groups, automate summaries in consulting, or use AI search in compliance. Headcount changes appear later through hiring freezes, smaller teams, vendor consolidation, and lower demand for contractors.

This lag can make AI displacement look smaller than it is. A company can reduce graduate hiring, leave vacancies unfilled, or avoid renewing temporary contracts without announcing AI layoffs. A bank can move work into a shared services unit. A software firm can expect the same sprint output from fewer junior developers. A media company can buy less freelance copy. None of those choices necessarily shows up as a dramatic mass layoff, yet all reduce labor demand for particular workers.

The Anthropic research on observed exposure adds another useful idea: AI’s theoretical capability is broader than actual use. Many tasks can be accelerated by a large language model, but fewer tasks are currently automated inside professional settings with enough reliability and integration to change hiring. This gap explains the uneven evidence. Some workers see little change. Others are already in markets where entry-level or routine work has become harder to sell.

Automation pressure also depends on how much tacit knowledge the task requires. A contract summary can be generated quickly, but deciding whether a clause is acceptable for a regulated customer requires commercial judgment, legal context, and risk tolerance. A model can draft code, but production deployment demands architecture, testing, security, maintainability, and accountability. A chatbot can answer common questions, but complaints, safety issues, refunds, fraud, and reputational risk need human handling.

Low-trust environments slow substitution. Health care, aviation, nuclear energy, finance, law, defense, child services, and public administration require documentation, auditability, and responsibility. AI may speed parts of these fields, but full substitution faces legal and professional barriers. In lower-risk settings, substitution can move faster because errors are cheaper and customer expectations are lower.

Workers most exposed to losses are those whose value is tied to producing standardized output without owning the customer relationship, system design, institutional decision, or accountability chain. That group includes many entry-level knowledge workers. It also includes outsourced service workers and contractors whose tasks are already packaged, measured, and priced as outputs. AI does not need to be perfect to affect them. It only needs to be cheaper, faster, and good enough for a defined slice of work.

Shortages Grow Where AI Raises the Skill Floor

The job-shortage argument starts with a different observation: AI adoption creates demand for people who can combine tool fluency with domain knowledge. The PwC 2026 barometer analyzed more than one billion job ads and described a two-path labor market in which employers reward human skills such as judgment, leadership, creativity, and communication. Its findings also pointed to faster headcount growth and wage gains among companies more able to use AI.

Shortage claims are credible when they describe specific skills, not a vague lack of “AI people.” Employers need machine learning engineers, data engineers, cloud architects, cybersecurity staff, governance specialists, privacy lawyers, product managers, domain analysts, trainers, auditors, prompt workflow designers, and managers who can redesign processes without creating legal, operational, or reputational problems. They also need frontline workers who can use AI outputs without surrendering professional judgment.

The ManpowerGroup 2026 survey of more than 39,000 employers across 41 countries reported that 72% had difficulty filling jobs and that AI capabilities had become the hardest-to-find skill category globally. That is a shortage claim, but it is also a diagnosis of mismatch. Employers may have applicants, yet not enough applicants with the right mix of AI literacy, work experience, domain understanding, and communication skill.

Shortages can worsen when companies cut the junior positions that feed the senior pipeline. Entry-level work has always done more than produce output. It teaches industry language, customer handling, quality norms, escalation judgment, documentation habits, team routines, and institutional standards. If AI removes too much routine work from juniors, organizations may save money in the short run and weaken their own talent supply later.

This problem appears in technology, accounting, consulting, media, finance, law, and insurance. Apprenticeship depends on low-risk tasks that give beginners exposure to real work. If software absorbs those tasks, organizations must design new learning pathways. Shadowing, simulation, supervised AI review, project rotations, and structured practice can replace some lost learning, but they cost time and management attention. Companies that skip that design may complain about senior shortages after narrowing their own entry gates.

The Indeed Hiring Lab reported in January 2026 that job postings were flat or declining overall, yet pockets of hiring growth were tied to AI skills. That pattern is exactly what a shortage economy looks like inside a soft hiring market. Openings do not rise everywhere. They rise where firms believe AI capability can protect margins, speed product delivery, reduce costs, or open new revenue.

AI can also generate shortages outside software. Data-center construction needs electricians, power engineers, cooling specialists, grid planners, permitting experts, environmental analysts, and facilities operators. New Space Economy’s coverage of AI workload types shows how training, inference, simulation, robotics, secure analytics, and satellite data processing place different demands on compute infrastructure. Those demands become labor demands when firms need people to site, build, operate, secure, and finance the physical systems behind AI.

Entry-Level Work Becomes the Stress Test

The entry-level labor market is where the job-loss and job-shortage stories collide most directly. Junior work often contains the tasks easiest to automate: summarizing documents, cleaning data, drafting standard text, preparing slides, writing test code, reviewing simple tickets, tagging content, conducting basic research, and producing first-pass analysis. At the same time, employers increasingly want new hires who can already use AI, check AI output, communicate with clients, and make sound judgments.

That creates a difficult bargain for graduates. AI can make them more productive, but employers may expect them to arrive with abilities that earlier generations developed through paid work. If entry-level posts demand mid-career judgment, many qualified beginners will appear underqualified. The shortage then shows up as an experience shortage rather than a headcount shortage.

The risk is stronger in fields where junior output has been treated as billable production. Consulting analysts, law associates, junior coders, marketing coordinators, insurance analysts, and financial analysts often learn by doing the tasks AI now accelerates. Employers may still need senior people to interpret results, manage clients, and take responsibility, but the junior ladder narrows.

A firm that cuts junior hiring can look efficient for two or three years. Work gets done. Margins improve. Senior staff use AI assistants. Then a promotion gap appears. Middle managers become harder to find. Senior employees carry too much review work. The company complains that candidates lack business judgment, but the company removed the work that once built that judgment.

Education systems cannot solve this alone. Universities can teach AI literacy, statistics, ethics, communication, and domain reasoning. They can use simulated clients and project-based learning. Yet real professional judgment comes from exposure to live constraints: unhappy customers, messy data, missed deadlines, unclear instructions, budget limits, regulatory ambiguity, and managers with competing priorities. Employers still own much of that learning environment.

The labor-market effect may vary by occupation. In software, code assistants can make strong juniors more capable, but they can also reduce demand for those who only perform narrow coding tasks. In law, AI can speed document review, but lawyers still need reasoning, negotiation, client trust, and court procedure. In health care, AI can support triage and imaging workflows, but licensing and patient responsibility keep humans central. In customer service, chatbots can absorb high-volume routine contacts, leaving human staff with harder cases and more stress.

The policy issue is not simply whether young people learn AI tools. It is whether the economy preserves enough paid routes into skilled work. Apprenticeship models, employer tax incentives for training, public-sector internships, professional licensing reform, wage subsidies for early-career workers, and clearer occupational skill standards can all help. Each option carries costs, but the alternative is a labor market with too few bridges between education and responsibility.

New Space Economy’s AI workforce analysis framed the central issue as task automation and job augmentation rather than simple mass job elimination. That framing is useful for entry-level work because the danger is not always a vanished occupation. The danger is a broken learning sequence.

Productivity Gains Can Produce Both Layoffs and Hiring

AI productivity gains do not dictate one employment outcome. The result depends on demand, pricing, competition, regulation, management choices, and how quickly organizations redesign work. A productivity gain can reduce headcount if the company serves a fixed market with little room to grow. The same gain can raise hiring if lower costs expand demand, improve service, open a new product line, or let the company compete for customers it could not reach before.

A call center provides a simple case. If AI handles 40% of routine contacts and total customer demand is stable, the employer may need fewer agents. If AI reduces wait times, supports new languages, improves retention, and makes support affordable for more customers, the employer may redeploy agents to complex cases, sales support, onboarding, fraud, or account management. The technology is the same. The market response differs.

Software gives another case. Coding assistants can reduce the time needed to generate standard code. A company with a fixed product roadmap might reduce junior hiring. A company with a backlog of profitable features might ship more products and hire more product managers, designers, security engineers, sales engineers, support staff, and customer success teams. Productivity removes one bottleneck and reveals another.

Manufacturing, robotics, and physical infrastructure add another layer. AI can speed design, simulation, maintenance, and quality control. It can also create demand for technicians who install sensors, maintain automated equipment, manage safety procedures, and adjust processes on the shop floor. Jeff Bezos’s new AI startup, Prometheus, has been described as focused on faster physical manufacturing, which points to the same demand logic: if AI reduces the cost of turning ideas into products, the economy may need more people to test, build, sell, maintain, and regulate those products.

The Stanford AI Index reported that generative AI reached 53% population adoption within three years and that organizational adoption rose sharply. Broad adoption increases the chance of both labor substitution and new work. More users mean more AI-assisted output, more integration challenges, more policy disputes, more vendor markets, more training needs, and more security concerns.

The productivity question also has a distribution problem. Gains may accrue to customers through lower prices, workers through higher wages, firms through higher profits, or investors through higher valuations. The outcome depends on bargaining power, competition, labor institutions, regulation, and skill scarcity. If AI raises output per worker but labor markets are weak, wage gains may be limited. If AI-literate workers are scarce, they may capture more of the value.

The World Economic Forum forecast projected 170 million jobs created and 92 million displaced by 2030, for a net increase of 78 million jobs. Forecasts at that scale should be read cautiously because definitions, employer expectations, macroeconomic conditions, and technology adoption rates can shift. The direction of the forecast still supports a core point: job creation and job destruction can be part of the same AI transition.

The productivity path is shaped by management. Firms that use AI only to cut labor may achieve quick savings but lose training capacity, customer trust, and innovation paths. Firms that use AI to redesign services may hire differently, demand stronger skills, and build new internal markets for training. Public policy can influence those choices through procurement rules, tax treatment, workforce programs, transparency requirements, and competition policy.

The Skills Gap Is a Demand-Side Problem Too

Skills-gap language often blames workers for not keeping up. That is incomplete. A shortage appears when employers, schools, governments, and technology vendors fail to align training with actual work. The gap can come from outdated curricula, weak employer training, unrealistic job ads, poor credential systems, low wages, geographic mismatch, immigration barriers, licensing rules, weak childcare support, or a hiring process that filters out people who could learn quickly.

AI intensifies this mismatch because skill needs change at the level of tasks. A marketing worker may need AI-assisted research, brand judgment, data privacy awareness, and campaign analytics. A public servant may need policy reasoning, procurement literacy, data governance, and tool evaluation. A nurse may need clinical judgment plus comfort with AI-assisted triage. A satellite analyst may need geospatial knowledge, machine learning awareness, and data-quality skepticism.

Some employers create their own shortage by asking for too much. Job ads may demand years of experience with tools released only recently. Entry-level positions may ask for strategy, leadership, machine learning, customer management, and industry knowledge. Automated hiring filters can reject candidates with adjacent skills. The market then produces a paradox: workers cannot get experience because jobs demand experience, and employers cannot find experienced workers because they did not train enough beginners.

The Microsoft Work Trend Index described new work patterns around people managing AI agents, designing workflows, and building multi-agent systems. Those tasks require more than tool use. They require process design, accountability, escalation rules, and understanding of failure modes. A worker who can write a prompt is useful. A worker who can redesign a claims workflow, test outputs, document controls, and manage exceptions is harder to find.

Skills shortages also appear in sectors that do not look like AI sectors. Energy utilities need planners for data-center loads. Local governments need permitting staff. Universities need AI governance and assessment experts. Courts need rules for evidence and practice. Hospitals need clinical AI oversight. Insurers need model-risk analysts. Defense organizations need autonomy, cyber, and procurement specialists. The labor shortage is often in the layer that connects AI tools to existing institutions.

New Space Economy’s article on the Canadian AI strategy reported national targets tied to added economic growth, AI-related jobs, youth work placements, wider business adoption, and a public AI supercomputer. That example shows how governments frame AI as an industrial, training, and infrastructure agenda rather than a software adoption issue alone.

The same lesson applies globally. Countries with strong universities, immigration systems, digital infrastructure, public-sector procurement capacity, and employer training programs may turn AI into higher-value work. Countries with weak infrastructure and limited training may see fewer benefits and more inequality. The IMF’s exposure analysis made a similar point: advanced economies face higher near-term disruption because they have more cognitive-intensive employment, yet lower-income economies may lack the infrastructure and skilled workforce needed to capture gains.

Skills policy has to move beyond one-time retraining. Workers need modular learning, portable credentials, paid practice time, and employer recognition of demonstrated competence. Managers need training too because poor work redesign can waste AI investment and damage morale. The shortage story is not just that workers need skills. Organizations need the skill to use skills well.

AI Infrastructure Turns Digital Growth Into Physical Labor Demand

AI feels weightless to users because prompts arrive through phones, laptops, and cloud software. Behind that interface sits a physical supply chain: chips, servers, power systems, cooling systems, fiber networks, data centers, substations, backup power, land, water, permitting, security, and specialized maintenance. Growth in AI adoption can convert digital demand into construction, energy, manufacturing, logistics, and operations work.

The AI Index and related Stanford coverage pointed to large-scale AI data-center power capacity by 2026. New Space Economy’s article on AI compute requirements connects that physical buildout to workload type. Training large models, running inference at scale, processing satellite imagery, simulating engineering systems, and supporting secure government analytics can have different power, latency, reliability, and data-movement requirements.

That matters for labor. A region attracting AI data centers may see demand for construction trades, electrical workers, grid engineers, environmental consultants, facilities managers, network technicians, and local-government staff. These jobs are not always called AI jobs, yet they are part of the AI labor market. Shortages in these fields can slow deployment as surely as shortages in machine learning engineers.

The physical buildout also affects the space economy. Satellite operators use AI for Earth observation analytics, autonomy, anomaly detection, tasking, and onboard processing. New Space Economy’s coverage of NVIDIA space computing shows how accelerated computing is moving into orbital and ground-space applications. The workforce implications include spacecraft software, edge computing, thermal management, radiation-aware hardware, geospatial data science, and mission operations.

Speculative claims about orbital data centers need caution. New Space Economy’s AI market structure coverage treats the AI market as supplier-heavy, with chips, cloud services, and data-center capacity absorbing large budgets before many downstream applications settle into stable economics. The same caution applies to space-based compute. Some space workloads may benefit from edge processing close to sensors. Full-scale orbital cloud computing faces cost, reliability, regulatory, insurance, and launch constraints.

Even restrained adoption of AI in space systems can alter labor needs. Satellite firms may hire fewer people for manual image sorting and more people for data pipeline design. Ground-segment operators may need AI-assisted anomaly detection. Defense and civil agencies may need procurement staff who can evaluate algorithmic claims. Insurers and regulators may need staff who understand model failure, cyber risk, and accountability.

Infrastructure also creates geographic mismatch. AI jobs may cluster near technology hubs, data-center regions, universities, defense procurement centers, and cloud markets. Workers displaced from routine office roles may not live near these growth nodes or have the credentials needed to enter them. Remote work can reduce some friction, but construction, energy, secure operations, health care, and many industrial tasks remain place-based.

The labor shortage story becomes more persuasive when AI demand crosses into these physical systems. It is easier to scale software than to scale electricians, power transformers, grid approvals, cooling infrastructure, launch capacity, secure facilities, and trained operators. AI can help design those systems, but humans still have to build, inspect, maintain, finance, insure, and govern them.

Global Effects Differ by Income, Sector, and Governance

AI’s labor impact will not land evenly. Advanced economies have more workers in exposed cognitive occupations, more digital infrastructure, and more capital to adopt AI tools. Emerging markets may have lower immediate exposure in some formal sectors but face risk if outsourced service work, call centers, translation, coding, and back-office operations become easier to automate or reshore.

The IMF’s estimate that advanced economies face about 60% job exposure, compared with 40% in emerging markets and 26% in low-income countries, points to this uneven pattern. Higher exposure can mean higher risk and higher opportunity. Countries with strong infrastructure and training systems may raise productivity. Countries without those assets may see fewer gains and weaker bargaining power in global service markets.

Sector differences may be just as large as national differences. Finance, software, professional services, media, marketing, insurance, and administration have many text-heavy workflows. Health care, education, logistics, manufacturing, agriculture, construction, energy, and public safety combine digital tasks with physical presence, regulation, trust, and local constraints. AI can affect each sector, but the mechanism differs.

Governance influences the pace of adoption. Privacy rules can slow data use. Liability rules can limit automation in high-risk settings. Public procurement can either reward safe adoption or buy weak systems. Labor law can shape consultation and retraining duties. Immigration policy can ease or worsen shortages in high-demand fields. Competition policy can determine whether productivity gains concentrate in a few firms or spread through the market.

Unionized sectors may also bargain over AI differently. Some unions will resist headcount reduction, surveillance, deskilling, and unsafe automation. Others may support AI that reduces hazardous work, paperwork, or excessive workloads. The dividing line is often control: workers are more likely to accept tools that improve their work and less likely to accept systems used mainly to monitor, replace, or intensify labor.

Public-sector adoption deserves separate attention. Governments face pressure to use AI for service delivery, fraud detection, procurement, health administration, tax processing, translation, and citizen support. They also carry higher accountability duties. A bad AI decision in public benefits, immigration, policing, child welfare, or taxation can harm rights and trust. That means public agencies may need more AI governance staff, not fewer public servants.

Education systems face a related challenge. AI can help tutoring, feedback, accessibility, and lesson planning. It can also disrupt assessment, writing instruction, and student skill signaling. Employers may discount credentials if they cannot tell whether graduates can think independently. Schools then need new ways to prove competence: oral exams, supervised projects, work-integrated learning, portfolios, and assessments that test judgment rather than output alone.

The global labor story is less a race between humans and machines than a race between institutional adaptation and technology adoption. Where institutions adapt slowly, AI can widen inequality, reduce early-career access, and concentrate gains. Where institutions adapt well, AI can raise productivity, reduce drudge work, improve services, and create new skilled work.

Management Choices Decide the Near-Term Outcome

Executives often discuss AI as if the technology itself decides the employment outcome. In practice, management choices determine whether AI becomes a layoff machine, a productivity tool, a training accelerator, or a source of operational chaos. The same tool can produce different labor outcomes depending on incentives and governance.

A cost-cutting strategy starts with a narrow question: how many people can be removed from the workflow? That strategy searches for tasks to automate, vendors to consolidate, and layers to reduce. It may improve margins quickly. It can also overload remaining staff, reduce service quality, weaken training pipelines, and make the organization dependent on tools it does not understand.

A capability strategy asks a different question: what work becomes possible if people have better tools? That strategy redesigns services, trains workers, measures quality, preserves accountability, and builds new paths from junior work to senior responsibility. It may still reduce headcount in some areas. It can also expand hiring in sales, product, operations, compliance, security, and customer support when new demand appears.

Both strategies can coexist. A company may automate routine support and hire more implementation specialists. A law firm may reduce document-review hours and hire more data-governance lawyers. A hospital may use AI for imaging support and hire clinical informatics staff. A government agency may automate form checking and need more appeals officers, privacy staff, and service designers.

The following table shows how a management decision changes the labor outcome.

Workforce planning also needs better measurement. Counting AI tool licenses says little. Managers need to know which tasks changed, which jobs lost learning value, which outputs improved, which errors increased, which workers gained time, and which customers received worse service. A firm that cannot measure these effects may mistake short-term speed for long-term performance.

Trust is another management variable. Workers may resist AI when they expect surveillance, layoffs, deskilling, or unfair evaluation. They may support AI when it reduces low-value paperwork, improves safety, speeds learning, or gives them more control. Involving workers in workflow design can reveal failure points that software vendors miss. It can also reduce the rumor cycle that often surrounds automation.

Procurement matters as well. Buying an AI tool without changing data governance, training, cybersecurity, and accountability can create hidden labor costs. Staff must clean data, rewrite processes, fix errors, answer complaints, and manage vendors. Poorly managed AI can increase work rather than reduce it, but that extra work may be invisible in early business cases.

The near-term employment result will come from thousands of these local decisions. Some employers will use AI to cut. Some will use AI to grow. Many will do both. The public debate may keep searching for a single answer, but the labor market will deliver mixed outcomes.

Policy Needs to Protect Mobility Rather Than Freeze Jobs

Policy cannot freeze the labor market in its pre-AI form. That would block productivity gains and make firms less competitive. Policy also cannot assume that market forces will automatically move displaced workers into better jobs. The policy goal should be mobility: giving workers practical routes from declining task bundles into growing ones.

That starts with better labor-market data. Governments should track AI-related vacancies, entry-level hiring, wage changes, displacement, training access, and regional concentration. Job title data is not enough because AI changes tasks inside occupations. Labor statistics agencies, universities, and private job platforms can help build task-level measures without exposing personal data.

Training policy should reward work-linked learning. Short courses can help, but workers need applied projects, recognized credentials, and chances to practice with real workflows. Public funding should favor programs tied to employer demand, wage outcomes, and transparent skill standards. Training should include AI literacy, data judgment, privacy, cybersecurity basics, communication, and domain-specific application.

Income support still matters. Workers displaced by AI may need wage insurance, unemployment benefits, relocation support, childcare help, disability accommodations, and paid training leave. The goal is not to subsidize every job indefinitely. The goal is to prevent a short-term layoff from becoming long-term exclusion.

Competition policy can shape outcomes. If AI productivity gains concentrate inside a few dominant firms, job losses may combine with weaker wage bargaining and reduced market entry. If AI tools diffuse through small and medium-sized businesses, productivity gains may create more local demand. Open standards, fair cloud access, data portability, and procurement rules can influence that diffusion.

Education policy needs revision, not panic. Schools should teach students to use AI without losing independent reasoning. Assessment should value explanation, judgment, source evaluation, collaboration, and oral defense of work. Work-integrated learning should expand because employers increasingly need evidence that graduates can perform under real constraints.

Immigration policy also affects shortages. Countries competing for AI engineers, semiconductor experts, health technologists, power engineers, and cybersecurity staff may need faster pathways for scarce skills. At the same time, immigration cannot substitute for domestic training. A country that imports all senior talent and neglects entry-level pathways may still face a fragile labor market.

Public procurement can set norms. Governments buy software, cloud services, health systems, defense tools, education platforms, and citizen-service technology. They can require human oversight, documentation, accessibility, data protection, audit rights, and workforce plans. That can push vendors and agencies toward better implementation.

The most useful policy question is not whether AI should be slowed or accelerated in the abstract. It is which institutions help workers move from threatened tasks to higher-value tasks, and which institutions allow gains to concentrate. AI creates a labor shock. Policy decides how much of that shock becomes mobility and how much becomes exclusion.

The Most Likely 2026 Answer Is Uneven Displacement and Uneven Shortage

By June 2026, the strongest evidence supported neither a simple jobs apocalypse nor a frictionless boom. The labor market showed uneven displacement pressure, uneven shortage pressure, and rising demand for workers who can combine AI fluency with human judgment and sector knowledge.

Some job losses are real. Employers have already connected AI to efficiency programs, hiring restraint, contractor reductions, and smaller teams. The most exposed workers are those tied to routine information tasks with weak customer ownership and limited decision authority. Entry-level workers face added pressure because many traditional learning tasks are now machine-assisted.

Some shortages are also real. Employers are struggling to find people who can build, deploy, govern, secure, audit, and manage AI systems. They need workers who understand the technology and the surrounding business, legal, ethical, customer, and operational context. Infrastructure demand adds pressure in energy, construction, data centers, chips, networks, and space-enabled analytics.

The two trends feed each other. Cutting junior roles can deepen shortages later. Raising skill requirements can make job losses harsher for workers without paid learning time. Expanding AI infrastructure can create jobs in places that differ from the places losing routine office work. Productivity gains can fund growth or fund headcount cuts, depending on market demand and managerial intent.

The central labor-market risk is polarization. High-skill workers who can use AI to increase judgment, creativity, management capacity, and technical reach may earn more. Workers whose tasks are standardized and easy to check may face weaker bargaining power. Workers outside strong training systems may be asked to adapt without receiving time, tools, or credentials.

A better outcome is possible but not automatic. It requires employers to preserve learning pathways, governments to support mobility, schools to change assessment and work placement, and workers to develop both AI fluency and domain judgment. The most valuable worker in an AI-rich workplace may not be the person who uses the newest tool. It may be the person who knows when the tool is wrong, what the business needs, who bears responsibility, and how to turn faster output into trusted work.

The job-loss and job-shortage perspectives are not rival prophecies. They are two views of the same transition. AI reduces demand for some tasks and raises demand for others. It weakens some career ladders and creates new ones. It can save labor in one corner of the economy and expose severe talent scarcity in another. The outcome will depend less on slogans about automation and more on how societies rebuild the connection between productivity, training, wages, and trust.

Summary

The debate over AI job losses and job shortages becomes clearer when jobs are broken into tasks. AI can automate routine information work, lower hiring in exposed occupations, and narrow entry-level pathways. It can also expand demand for scarce skills, create new infrastructure work, and raise the value of judgment, domain knowledge, communication, and governance.

Evidence available by June 2026 points to a mixed labor market. The IMF, ILO, Anthropic, and labor-market data support concern about displacement and weaker demand for some tasks. PwC, ManpowerGroup, Indeed, Stanford HAI, and employer behavior support the shortage case, particularly where AI adoption requires rare combinations of technical skill, business judgment, and institutional trust.

The most important policy and management choice is whether AI becomes a substitute for people or a system that raises what people can do. Some substitution is unavoidable. The avoidable damage comes from cutting training pipelines, treating workers as costs only, and failing to build practical routes into the jobs AI creates or expands.

Appendix: Useful Books Available on Amazon

• Co-Intelligence
• Power and Progress
• Prediction Machines
• Human Compatible
• The Coming Wave

Appendix: Top Questions Answered in This Article

Will AI Cause Mass Unemployment?

AI may reduce employment in some tasks and occupations, but current evidence does not support one universal outcome. Broad employment can keep growing even as specific entry-level, routine, or back-office work weakens. The more accurate risk is uneven displacement combined with weaker access to some career ladders.

Why Can AI Create Job Shortages?

AI creates shortages when it expands what organizations can attempt faster than they can hire qualified people. Firms need workers who combine AI fluency with domain knowledge, customer judgment, data governance, cybersecurity, and operational design. Those combinations are harder to produce than basic tool familiarity.

Which Workers Face the Highest Displacement Risk?

Workers face higher risk when their main tasks are repeatable, text-heavy, measurable, and easy to check. Routine support, basic drafting, standard coding, simple data processing, and some contractor work are more exposed. Risk falls when work requires trust, accountability, physical presence, regulation, negotiation, or complex judgment.

Why Are Entry-Level Jobs Under Pressure?

Entry-level jobs often include the tasks AI can perform or accelerate. Those tasks also teach beginners how professional work functions. If employers remove junior work without replacing the learning pathway, they may reduce costs now and create shortages of experienced workers later.

Does AI Always Raise Productivity?

AI can raise productivity when it fits the workflow, uses reliable data, and has human review. Poor implementation can create rework, compliance problems, customer distrust, and hidden labor costs. Productivity gains depend on management, measurement, training, and demand for the improved service.

Can AI Help Workers Rather Than Replace Them?

AI can help workers by reducing repetitive work, speeding analysis, supporting drafting, improving accessibility, and widening access to knowledge. The benefit depends on whether workers have control, training, and time to use tools well. AI used mainly for surveillance or headcount reduction produces a different outcome.

Why Do Employers Still Report Talent Shortages?

Employers report shortages because AI adoption requires combined skills that are scarce. They need technical fluency, sector knowledge, judgment, communication, and governance ability in the same person or team. Hiring systems can worsen the shortage by demanding experience that beginners have had little chance to gain.

What Should Governments Do About AI and Jobs?

Governments should support mobility rather than freeze old job structures. Useful tools include labor-market data, paid training, wage support, work-integrated education, credential reform, fair procurement, competition policy, and targeted immigration for scarce skills. The goal is to help workers move into growing work.

How Does AI Infrastructure Affect Employment?

AI infrastructure creates labor demand in data centers, power systems, cooling, networking, construction, permitting, cybersecurity, and facilities operations. These jobs may not carry AI titles, but they support AI adoption. Shortages in physical infrastructure can slow digital deployment.

What Is the Best Forecast for AI’s Labor Impact?

The best forecast is uneven change. Some jobs will shrink, some will grow, and many will be redesigned. Workers with AI fluency, domain judgment, and communication skill may gain bargaining power. Workers tied to routine task production may face weaker hiring and wage pressure.

Appendix: Glossary of Key Terms

Agentic AI

Agentic AI refers to systems that can plan and carry out sequences of tasks with less step-by-step human instruction. In labor markets, these systems matter because they can move beyond single responses into workflow execution, software actions, scheduling, data retrieval, and multi-step business processes.

Automation

Automation means using technology to perform a task that a person previously performed. In AI labor discussions, automation usually refers to software handling routine cognitive work such as drafting, classifying, summarizing, coding, routing, or checking standard information.

Augmentation

Augmentation means AI supports a worker rather than replacing the worker. Examples include drafting a document for human review, suggesting code, summarizing customer history, or flagging anomalies. The worker remains responsible for judgment, quality, context, and final decisions.

Exposure

Exposure measures how much of a job’s task content could be affected by AI. It does not automatically mean job loss. A highly exposed job can be improved, redesigned, partially automated, or reduced depending on tool quality, cost, regulation, and management choices.

Job Shortage

A job shortage occurs when employers cannot find enough qualified workers for available work at prevailing conditions. In AI markets, shortages often involve combined skills, such as data knowledge plus business judgment, or machine learning skill plus regulatory understanding.

Labor Demand

Labor demand is the amount and type of work employers want to buy through hiring, contracting, or internal assignment. AI can reduce labor demand for some tasks and increase it for others by changing costs, productivity, customer expectations, and business models.

Large Language Model

A large language model is an AI system trained on large amounts of text and other data to generate, transform, classify, or summarize language. These models affect work because many office tasks involve writing, reading, searching, explaining, and decision support.

Reskilling

Reskilling means training workers for substantially different tasks or occupations. AI reskilling may include tool use, data judgment, workflow redesign, cybersecurity awareness, privacy rules, and domain-specific application rather than coding alone.

Skills Gap

A skills gap exists when workers’ abilities do not match employer needs. In AI adoption, the gap can come from worker training, unrealistic job ads, weak employer development, poor credential systems, or changing requirements inside existing occupations.

Task Redesign

Task redesign means changing how work is divided between people, software, teams, and vendors. AI task redesign can remove routine steps, add review duties, create escalation rules, change training paths, and alter what employers expect from each job.