The forecast has once again arrived ahead of the event. Fortunately, the forecast has a flexible return ticket.
A September McKinsey chart reports that 32 percent of respondents in 2025 expected AI-related reductions in their organizations’ headcount. In 2026, 14 percent reported reductions over the preceding twelve months, while 39 percent expected reductions in the coming year. These are responses to different questions about organizations, not percentages of all jobs eliminated. They should not be treated as a matched longitudinal record of the same firms without further evidence.1
Three numbers, three questions
The difference is not a pedantic objection. A company can reduce hiring without reducing total headcount. It can change tasks without eliminating roles. It can lose employees for several reasons and find it difficult to attribute the change to one technology.
When these distinctions disappear, a chart about organizational expectations becomes a statement that a specified share of humanity is about to lose its job. The dramatic version travels faster. It has also changed the subject.
The productivity demonstration has not issued a termination notice
A model completing a task shows that the task can be performed under particular conditions. The path from that capability to employment effects includes organizational decisions, demand, costs, quality requirements, and the other work included in a job.
Some tasks may become cheaper, allowing more of them to be done. Some roles may change. Some work may be consolidated. Some positions may disappear. The balance can differ across industries and periods.
A person affected by a real job loss gains little from being told the aggregate story is complicated. A person making policy gains little from being told every demonstration establishes an economy-wide conclusion. Both deserve better evidence.
The forecast has incentives too
Write down the deadline, the workers covered, and the assumptions connecting a new capability to fewer jobs. Keep that version somewhere the next forecast cannot overwrite it.
The opposite camp has its own temptation: use slower-than-expected change as proof that no substantial change is coming. That is equally premature. Delays can reflect implementation friction rather than a permanent limit.
The quieter changes deserve attention
Total headcount is only one part of the employment story. The composition of work, opportunities for new entrants, supervision demands, compensation, and bargaining power may change without a dramatic net reduction.
These are possibilities to investigate, not effects established by the cited chart. They matter because a narrow metric can miss changes that workers experience directly.
Distinguishing those outcomes takes detail about what changed inside the organization. The word “jobs” has been asked to do too much summarizing.
The useful forecast is revisable
A good forecast should make uncertainty visible and update when evidence changes. Revising an estimate is not necessarily failure. Pretending the old estimate never existed is less helpful.
The McKinsey comparison supplies a useful pause in the argument. Reported reductions were less widespread than the earlier expectations, while expectations for the next period remained substantial.1 That is a reason to keep investigating, not to declare the debate over.
The future may change the labor market profoundly. It should still be required to show up in the data.