The forecasts disagree, and that is the point
Anyone giving you one confident number is selling something.
9%, 14%, 47%, 80%, 300 million jobs. The serious estimates of AI's effect on work span an enormous range, because they measure different things (whole jobs vs tasks, exposure vs automation) and make different assumptions about how fast and how far adoption goes. The honest reading is not to pick your favourite number but to hold the spread, and to plan for a range rather than a point.
What the evidence shows
Frey & Osborne: about 47% of US jobs at high risk (occupation-level).
OECD task-based: about 9% at high risk, with another third significantly changed.
OpenAI exposure study: about 80% of workers with at least 10% of tasks affected.
Goldman Sachs: the equivalent of 300 million full-time jobs exposed globally.
Disagreement is not the same as 'nobody knows anything'. The studies broadly agree on direction (knowledge tasks exposed, social and physical tasks less so) even where they differ wildly on magnitude. Use the agreement; distrust the precision.
Build a career that holds up across the range, not one that only works if the optimistic number is right.