AI and Work: What the Evidence Actually Shows So Far

When people ask whether AI is coming for their jobs, they usually want a number. The research offers many numbers, but they measure different things, and reading them carefully matters more than repeating the scariest one.

How many jobs are "exposed"

The most-cited figure comes from the International Monetary Fund, which estimated in January 2024 that about 40 percent of jobs worldwide are exposed to AI — roughly 60 percent in advanced economies, 40 percent in emerging markets, and 26 percent in low-income countries. The International Labour Organization, using a refined 2025 index, put the share of jobs with meaningful generative-AI exposure at about one in four globally, and stressed that its findings point toward jobs being transformed rather than eliminated.

It is worth being precise about what "exposed" means. It describes tasks a model could plausibly touch — not jobs that will disappear. The IMF itself split its 40 percent roughly in half: some exposed jobs may be made more productive, others may face pressure. Exposure is a map of where change could land, not a forecast of losses.

Productivity: real gains, unevenly shared

On productivity, the early evidence is more concrete because it comes from controlled studies. A widely cited field study of more than 5,000 customer-support agents found that access to a generative-AI assistant raised issues resolved per hour by about 14 percent on average — but roughly 34 percent for novices, with little measurable gain for the most experienced workers. In software, a GitHub–Accenture randomized trial reported task-completion time falling substantially with an AI coding assistant, alongside modest gains in success rates.

A consistent pattern runs through these studies: the largest measured gains tend to go to less-experienced workers, narrowing the gap with veterans. Whether that dynamic raises the value of entry-level roles or reduces the need for them is exactly what the labor-market data is now being watched to answer.

Displacement, augmentation, and what the labor data shows

Here the findings diverge most. The World Economic Forum's 2025 Future of Jobs report projected 170 million jobs created and 92 million displaced by 2030 — a net gain, but with heavy churn as skill demands shift. Those are employer surveys and projections, not observations.

When researchers look at what has actually happened, the picture is calmer and more contested. The Budget Lab at Yale, tracking U.S. employment data since ChatGPT's release, has repeatedly found that broad measures of AI exposure show no clear, economy-wide link to changes in employment or unemployment so far — describing the picture as stability, not major disruption, while cautioning that the data is a snapshot, not a prediction.

A more specific signal comes from Stanford's Digital Economy Lab. Its 2025 study of payroll records reported that early-career workers (ages 22–25) in the most AI-exposed occupations saw a relative employment decline of roughly 13 to 16 percent since generative AI became widespread, concentrated in fields where AI automates rather than augments — while older and less-exposed workers held steady. The authors framed this as an early signal to watch, not a settled trend, and other economists note that a soft hiring market and other factors are hard to fully separate out.

Wages and who is most affected

Wage effects are the least settled of all. The OECD finds that AI is reshaping which skills employers value — raising demand for management and business skills even in highly exposed roles — and that most exposed workers will not need specialized AI expertise. Early data has shown higher pay associated with more AI-exposed jobs, but researchers caution this may reflect who holds those jobs rather than AI raising wages directly.

On who is most affected, several sources converge. Clerical work shows the highest exposure, and the ILO reports that women's jobs are more exposed than men's, largely because women are over-represented in clerical roles — with the highest-risk jobs making up about 9.6 percent of female employment in high-income countries versus 3.5 percent for men.

What remains genuinely uncertain

The honest summary is that the technology's potential reach is large and fairly well mapped, its measured productivity effects are real but uneven, and its realized effect on jobs and wages so far is small, mixed, and disputed. Estimates of AI's near-term boost to overall productivity range from modest to transformative; one influential economic analysis put the ten-year gain to total factor productivity at under one percent, while other institutions project far larger effects.

Data lags reality, adoption is still early, and separating AI's influence from interest rates, hiring cycles, and post-pandemic churn is genuinely difficult. The most defensible position today is not confidence in either direction, but attention: the signals worth watching — entry-level hiring, task composition, wage gaps — are visible, and the next few years of data will say far more than any single number can now.