Bosses predict AI will shrink jobs, employees predict their own jobs will grow — someone's math is off.

This explores whether what executives say AI will do to their workforce matches what actually happens to headcount and payroll. The short answer is that the collection can't yet settle this directly. No study here follows one group of executives' forecasts and then checks them against later payroll data. What the collection does have is several studies showing a gap between what people say and what gets measured. Together they suggest executives' predictions about jobs deserve some skepticism.

Start with the predictions themselves. An NBER survey of nearly 6,000 executives found they expect AI to cut employment by about 0.7% over three years. Employees at the same kinds of firms expect a 0.5% gain Do executives and employees agree on AI's job impact?. Both groups can't be right about the same companies, so at least one is badly calibrated. Both numbers are also small, which suggests that even the pessimistic view is a nudge to headcount, not mass layoffs.

The closest thing to a reality check comes from a survey of 750 executives. It found that the productivity gains executives *perceive* from AI are larger than what can be *measured*, likely because revenue lags behind day-to-day improvements. The labor data in that study showed workers being shifted into different roles rather than cut overall, concentrated in high-skill services and finance Do AI productivity gains feel larger than they actually measure?. That pattern matters for the payroll question. If AI mostly moves people around, a forecast of "fewer jobs" can be wrong about the total even while it's right that some roles disappear.

The less obvious lesson comes from outside economics. A study of AI competence found that people's self-ratings correlate almost not at all (r ≈ .055) with how well they actually perform Can self-ratings replace objective performance scores for AI competence?. Related work describes an "LLM Fallacy": people credit AI-assisted output to their own ability, which inflates how capable they believe they are How does AI-assisted work reshape how people see their own abilities?. Executive forecasts are also a form of self-report, about their own organizations rather than themselves. That makes it reasonable to expect the same distortion.

The workplace surveys in the collection share a further weakness: they are correlational, and they come from interested parties. Anthropic found that its heaviest Claude users are the most optimistic about their careers Does delegating work to AI actually damage worker skills?. Gallup found that support from managers goes along with better ratings of workplace culture Does manager support actually shape how employees experience AI at work?. Neither checks the reported sentiment against hard outcomes. There is a useful contrast from market research: AI personas built from real *behavioral* data predict A/B test results far better than stated preferences would Can behavior-based personas predict A/B test outcomes?. Applied to jobs, the takeaway is to trust payroll records and hiring data over what anyone says they plan to do. The collection is still missing a study that puts those two side by side.