AI Needs A Jobs Compact, Not A Panic Cycle
AI will not simply "take all jobs" or "create abundance" by itself. It will reorganize tasks, bargaining power, wages, training, hiring, and confidence. If workers are expected to absorb the shock alone, society will push back. The answer is not panic. It is a jobs compact.
There is a lazy way to talk about AI and jobs. It comes in two versions.
The first says AI will destroy work, hollow out the middle class, wipe out entry-level careers, and leave millions of people staring at a screen that has learned to do what they once did. The second says AI will create more jobs than it destroys, lift productivity, free workers from drudgery, and deliver abundance if governments simply stay out of the way.
Both versions are too clean for the world we actually live in.
AI is not a demon that automatically produces unemployment. It is already helping workers draft, code, translate, design, diagnose, search, summarize, simulate, and automate tasks that used to take hours. For many people it can become a power tool.
But AI is not an angel that automatically distributes prosperity either. It is built and deployed inside labour markets with unequal bargaining power, uneven training access, weak safety nets, platform concentration, and firms under pressure to cut costs. It can augment a worker. It can also replace a worker, reduce hiring, compress wages, deskill a task, or move the gains to shareholders while telling everyone else to reskill.
The honest question is not whether AI is good or bad for jobs. The honest question is: who gets the productivity, who carries the transition, and who has power when the work changes?
That is why AI needs a jobs compact, not a panic cycle.
Exposure is not destiny
The word "exposure" appears constantly in studies of AI and work. It measures how much of a job's task content could be affected by AI. It is useful, but it is also easily misunderstood.
Exposure does not mean elimination. A teacher, lawyer, accountant, journalist, clerk, software developer, designer, translator, customer-support worker, or analyst may be exposed to AI because many tasks in the job can be assisted by AI. That does not tell us whether the worker becomes more productive, loses hours, earns more, earns less, changes role, moves firm, or exits the occupation.
The International Labour Organization warned in April 2026 that AI exposure indicators should be treated as early warning signals, not forecasts carved in stone. They depend on static task lists, assumptions about adoption, economic conditions, institutional barriers, and differing definitions of what exposure means. To understand real labour impact, they must be paired with evidence on wages, employment, job transitions, and the institutions shaping adoption.
That distinction matters because bad policy can be built from dramatic numbers. If exposure is treated as inevitable replacement, the public conversation collapses into fear. If exposure is dismissed because unemployment has not yet surged, the public conversation collapses into denial.
The middle is where policy lives.
AI will change work unevenly. Some people will be helped. Some will be displaced. Some will see the easy parts of their jobs automated while the stressful human parts remain. Some will be asked to supervise systems they do not control. Some will become more productive without receiving more pay. Some will enter careers where the first rung of the ladder has quietly disappeared.
That last point deserves special attention.
The missing first rung
Many professions teach people by letting them do junior work. A young lawyer reviews documents. A junior analyst builds summaries. A trainee accountant reconciles records. A reporter drafts briefs. A designer prepares variations. A developer fixes small issues. A translator handles simpler text. A customer-support worker learns the product by answering routine questions.
AI is especially attractive to firms at precisely this level. Routine cognitive tasks are easier to automate than responsibility. That creates a quiet danger: not mass unemployment overnight, but fewer entry points.
If the first rung disappears, the ladder becomes decorative.
This is not only a problem for young workers. It is a problem for the whole economy. Senior workers do not appear from nowhere. Expertise is cultivated through low-status tasks, repetition, error, feedback, supervision, and gradual responsibility. If companies automate too much of the training ground, they may save money today and discover a talent drought tomorrow.
Society cannot treat this as an internal HR problem. Entry-level work is civic infrastructure. It is how people become taxpayers, consumers, professionals, parents, homeowners, community members, and future mentors. A labour market that blocks young people from stable beginnings produces anxiety that travels far beyond the office.
AI companies and employers should understand the feedback loop. They need consumers. Consumers need income. Income needs work. Work needs trust that effort still has a path. If AI adoption weakens that chain, the market for AI itself becomes politically and economically unstable.
The disruption may arrive before the dividend
In March 2026, the ILO and World Bank released a background study for the World Development Report 2026 examining generative AI exposure across 135 countries, covering around two-thirds of global employment. The central warning was not that every country faces identical automation risk. It was that the benefits and disruptions are likely to arrive unevenly.
Developing economies may have lower aggregate automation exposure than advanced economies because of their occupational structures, but they may still face comparable potential for task augmentation. The risk is that disruption can arrive faster than productivity gains, especially where digital gaps, skills shortages, limited connectivity, weak social protection, and institutional constraints prevent workers and firms from capturing the upside.
That is a crucial TGF point. AI is not only a labour-saving technology. It is also a divider if adoption capacity differs.
A worker in a rich country may use AI to become more productive, supported by training, good connectivity, legal rights, and a firm that can reorganize tasks. A worker in a poorer country may face competition from AI-enabled outsourcing, shrinking clerical work, or platform-driven wage pressure without receiving the tools, training, or infrastructure needed to benefit.
The same technology can be augmentation in one context and displacement in another.
The World Bank's 2026 development work takes the same balanced line. AI may help developing countries address market failures, improve credit access, fill skill gaps in education and health, and support small businesses. But the Bank also warns that AI can widen gaps between high- and lower-income countries because of compute, data, and skills requirements; automate tasks in ways that reduce jobs or wages; concentrate advantage among large technology companies; and reinforce bias or flawed decisions without safeguards.
That is not anti-AI. It is reality. A technology that can do both good and harm must be governed for the good.
Headline unemployment can hide social stress
The labour market does not always announce distress with one clean number.
The ILO's 2026 Employment and Social Trends work projects global unemployment at about 4.9 percent in 2026, a sign of resilience at the headline level. But it also warns that job quality has stalled, inequalities remain, young people continue to struggle, and AI plus trade uncertainty could worsen the outlook if risks materialize.
This is what policymakers often miss. A society can have stable unemployment and still be anxious. Hiring can slow before layoffs rise. Wages can stagnate while productivity improves. Entry-level postings can shrink while senior workers stay employed. Workers can keep jobs while losing autonomy. More people can work while fewer feel secure.
If AI affects employment mainly through reduced hiring, task compression, lower wage growth, contract work, and fewer career ladders, the political shock may arrive before the unemployment rate explains it.
That shock will not be irrational. It will be the lived experience of people who see the bargain changing.
What a jobs compact should contain
A jobs compact is not a promise that no job will ever change. That would be dishonest and impossible. A jobs compact is a promise that workers will not be abandoned while work changes.
It should have several parts.
First, notice. Workers should have meaningful advance warning when AI systems are being introduced in ways likely to alter job roles, performance measurement, staffing levels, or work intensity. Surprise automation breeds fear and resistance.
Second, consultation. Workers and their representatives should have a voice in how AI is deployed. The people who do the work often know where tools will help, where they will fail, and where management is using "AI" as a cover for ordinary cost cutting.
Third, training time. Reskilling cannot be treated as homework after exhaustion. If firms benefit from AI productivity, they should contribute paid time for workers to learn the tools and move into new roles.
Fourth, transition support. Wage insurance, placement services, portable benefits, public retraining funds, and unemployment systems designed for mid-career transition should be part of AI policy, not an afterthought.
Fifth, protection for entry-level pathways. Firms should be encouraged, and in some sectors required through public procurement or professional standards, to maintain training roles for young workers. Automation should not be allowed to burn the seed corn of future expertise.
Sixth, transparency. Companies should track and disclose, at least to regulators and workers, how AI deployment affects headcount, hiring, wages, work intensity, discrimination risk, and productivity distribution.
Seventh, public AI literacy. Citizens should not learn about AI only from corporate marketing or fear-driven headlines. Schools, libraries, unions, community colleges, and public broadcasters can help people understand what AI can do, where it fails, and how to use it responsibly.
Finally, shared gains. If AI raises productivity, the gains should show up not only in margins, valuations, and executive decks, but in wages, shorter workweeks where feasible, better services, lower prices, and public revenues.
That is how a society keeps consent.
Business has a stake in the compact
Some firms may hear "jobs compact" and think "cost." That is too narrow.
A frightened workforce adopts badly. A resentful workforce withholds knowledge. A public that believes AI is a device for replacing people will support harsher restrictions. Consumers without income do not buy products. Democracies that see technology as extraction will regulate through anger rather than design.
The smarter firm wants a stable transition because trust is an asset.
The World Economic Forum's Future of Jobs work projects large-scale labour churn by 2030, with 170 million new roles and 92 million displaced roles in its 2025 report, resulting in a projected net gain but also enormous transition. It also found that 63 percent of employers cited skills gaps as a key barrier and that 77 percent expected to upskill workers in response to AI, while 41 percent planned workforce reductions as AI automates tasks.
That is the compact in numbers. Employers need skills. Workers need security. Societies need productivity without despair.
If business treats AI transition as a private optimization exercise, politics will eventually treat it as a public harm. If business treats workers as partners in adoption, AI has a better chance of becoming a productivity story rather than a legitimacy crisis.
The state cannot outsource adjustment
Governments also have work to do.
They cannot simply tell people to "learn AI" while cutting education, underfunding public employment services, and allowing regional digital divides to harden. They cannot celebrate AI investment while ignoring the local labour market effects of data centres, automation, and procurement. They cannot buy AI into public services without asking what happens to the workers who understand those services.
A serious state should build labour-market observatories that track AI's real effects, not only theoretical exposure. It should connect unemployment insurance, skills programmes, community colleges, public employment agencies, and industrial policy. It should use public procurement to reward firms that train workers and disclose AI impacts. It should protect workers from algorithmic surveillance and automated performance scoring without appeal.
Most of all, it should remember that work is not only income. Work is identity, rhythm, social connection, dignity, bargaining power, and participation in the future.
A society that ignores that will misread the politics of AI.
No panic, no worship
AI will change work. It already is. Some warnings are exaggerated. Some assurances are naive. The correct response is neither panic nor worship.
Panic treats workers as doomed. Worship treats workers as friction. A jobs compact treats workers as citizens.
That distinction matters. Citizens can adapt when they are respected. They can learn new tools when given time and support. They can accept change when the gains are shared and the rules are visible. They can help improve AI systems because they understand real tasks better than a dashboard does.
But citizens will not quietly accept a future in which firms use AI to hollow out opportunity, concentrate gains, blame the displaced for failing to reskill fast enough, and then ask society to keep buying the products.
AI companies cannot exist in isolation. Employers cannot automate consumers out of purchasing power. Governments cannot outsource legitimacy to software. Even non-democratic systems cannot indefinitely damage the common good without paying a price in trust, stability, or performance.
The future of AI and work is not predetermined by the model. It will be shaped by institutions.
If those institutions are weak, AI will amplify insecurity. If they are strong, AI can amplify human capability.
That is the choice. Not jobs apocalypse. Not effortless abundance. A bargain.
And the bargain should be written before the shock arrives.
The Global Federation covers artificial intelligence as a civic question: not whether the technology advances, but whether the institutions around it keep enough power with the people it touches.