Before asking whether AI is eliminating jobs, separate three kinds of numbers: observed employment outcomes, estimates of how much work is exposed to AI, and employer forecasts about future hiring. They answer different questions. Mixing them turns a scenario that has not happened into an apparent record of layoffs.
This article is current to July 10, 2026. It does not invent five-year headcount percentages for individual occupations. Instead, it states what each public source supports and where the evidence stops.
What is observed: hiring has softened, but AI's independent effect is not settled
The OECD Employment Outlook 2026 reports an OECD-wide unemployment rate of 4.9% in May 2026 and an average employment rate of 72.1% in the first quarter. Those figures do not describe a collapsed labor market. They do show slower growth in employment and labor-force participation, along with increasing difficulty for young university graduates.
The limitation matters. The OECD says the relative deterioration for young graduates began before generative AI spread widely, which points to more complex causes. Local industry mix, the business cycle, trade shocks, and employer hiring practices move at the same time. A decline in postings for junior roles with high language-model exposure is therefore a signal to investigate, not a clean count of jobs lost because of AI.
For career decisions, national unemployment is not enough. The mix of tasks, career level, location, and adoption rate all matter. Two people with the same occupation title can face different exposure when one role centers on routine input and the other on client judgment, field work, or regulated accountability.
Exposure: one in four jobs is in scope, but exposure is not a layoff rate
The 2025 ILO–NASK occupational exposure study combines assessments of nearly 30,000 tasks with expert validation. It estimates that about one quarter of global employment falls within one of four generative-AI exposure gradients. The highest gradient represents 3.3% of global employment. Clerical occupations remain the most exposed, while exposure has also risen in highly digitized professional and technical work.
That does not mean one worker in four will lose a job. Most occupations contain tasks that still require human input, so the ILO identifies job transformation, rather than full replacement, as the most likely effect under its framework. Exposure is a measure of technical potential. Infrastructure, cost, regulation, organizational choices, and the need for accountable human work determine what is actually adopted.
| Measure | What it can answer | What it cannot answer |
|---|---|---|
| Employment and unemployment rates | Whether the overall labor market is strong now | How many layoffs AI caused |
| Task exposure | Which tasks AI may be technically capable of doing | The observed layoff rate for that occupation |
| Job-posting changes | How stated employer demand is moving | Whether AI alone caused the change |
| Employer forecasts | What surveyed firms expect through 2030 | Whether the forecast will occur exactly |
Forecasts: the 2030 totals are a scenario based on employer expectations
The WEF Future of Jobs Report 2025 combines responses from more than 1,000 employers with ILO employment data. Across technology, demographic change, the green transition, economic pressure, and geoeconomic change, it projects 170 million roles created and 92 million displaced by 2030, for a net increase of 78 million. Creation and displacement together represent structural churn equal to 22% of the formal jobs covered by its dataset.
These are not observed results, and they are not an estimate of AI's effect alone. They are modelled expectations across several macro trends. Saying that “AI has already removed 92 million jobs” would misstate the report. The useful individual-level signal is narrower: employers expect fast growth in AI, data, and cybersecurity skills while continuing to value analytical thinking, resilience, and collaboration.
A four-step check for your own role
- Break the role into tasks. Track weekly time spent on input and summarization, analysis and judgment, client or field work, and approval or accountability.
- Measure actual demand. Save eight to twelve weeks of comparable job-posting counts, required skills, and seniority. A single search result is not a trend.
- Separate automation from accountability. The ability to draft an answer is different from an organization's willingness to delegate the final decision. Include error cost, privacy, and regulatory responsibility.
- Update the strategy quarterly. Use tools to reduce exposed routine work, then document evidence of verification, decision-making, and stakeholder coordination where human responsibility remains.
Conclusion
As of July 2026, the public evidence does not support either “AI is eliminating every occupation” or “new jobs will automatically offset every loss.” OECD observations show resilient aggregate employment alongside uneven risks for young workers and regions. The ILO finds broad task exposure but considers transformation more likely than wholesale replacement. The WEF numbers are employer-informed forecasts across multiple macro trends, not a realized AI layoff count.
The practical response is to measure the task mix and real hiring demand for your role repeatedly, using the same method. These sources describe population-level direction; none can guarantee an individual's employment outcome.