Reading a forecast

Here are four real kinds of numbers about AI and jobs. They point in different directions.

• Exposure estimate (2023). Researchers from OpenAI and the University of Pennsylvania studied language models. About 80% of US workers could have at least 10% of their tasks affected. About 19% could see at least half their tasks affected. The authors said they were not predicting when.
• Employer survey (2025). The World Economic Forum projected 170 million jobs created and 92 million displaced globally by 2030, a net gain of 78 million.
• Company announcements (2026). Block cut 40% of its staff in February 2026, and its CEO pointed to AI. A job-cuts tracker counted about 113,000 cuts tied to AI by late August. Yet executives at Davos told Reuters that AI “would be used as an excuse” for cuts.
• Measured so far (2025 and 2026). Yale’s Budget Lab found no clear effect on the overall US labor market, most recently in May 2026. In March 2026, Anthropic researchers found “no systematic increase in unemployment” for the most exposed workers.

So which is right? Each answers a different question. exposure (show definition) tells you which tasks could be affected, not whether jobs disappear. A survey tells you what managers expect. An announcement tells you what a company wants you to hear. Labor data tells you what has happened, but only so far.

Ask four questions of any forecast:

1. Who published it, and what do they gain from the answer?
2. What did they measure: exposure, expectations, announcements, or real hiring and pay?
3. Over what time frame?
4. What do they mean by “affected”?