Nobody announces a stealth layoff. It assembles itself from small managerial acts: the backfill that quietly lapses, the junior scope absorbed by a license upgrade, the team of nine that becomes seven through attrition while output holds, and the all-hands where leadership repeats that AI is about augmentation. The manager in the middle executes every one of those acts. Usually without ever having been told, in words, that shrinking the team was the plan.
The ambient numbers explain the ambiguity. In the World Economic Forum's 2025 employer survey spanning more than 1,000 companies across 55 economies, 77 percent of employers said they plan to upskill their workforce in response to AI, and 41 percent said they plan to reduce headcount as AI automates tasks. Both intentions, frequently inside the same company. The gap between those two statements is where the millennial manager now works.
The gains are real, jagged, and junior-shaped
Three field studies define what a manager is actually redistributing. Brynjolfsson, Li and Raymond, studying 5,172 customer-support agents, found AI assistance raised productivity about 15 percent on average, with gains up to roughly 34 percent for novices and near zero for the most experienced; the tool effectively diffused the experts' patterns to beginners. Dell'Acqua and colleagues' experiment with 758 BCG consultants found GPT-4 users completed 12.2 percent more tasks about 25 percent faster with markedly higher quality, but only inside the technology's "jagged frontier"; on tasks just outside it, AI users made more errors, misled by fluent, wrong output. And Stanford's payroll-data work on early AI employment effects found workers aged 22 to 25 in the most AI-exposed occupations experienced a relative employment decline of about 16 percent, while experienced workers in the same occupations held steady, with the damage concentrated where AI automates rather than augments.
Put together: the productivity is genuine, the map of where it works is uneven and must be learned, and the tasks most cleanly automated are the ones juniors learn the job on. When a manager reroutes the routine work into the tool, she is not just saving hours. She is deciding whether her youngest people land on the automation side or the augmentation side of the line that the Stanford data says determines who keeps working.
Redistribution is the job; silence is the failure
What do workers themselves want moved? Stanford's WORKBank audit, crossing 1,500 workers' preferences with expert assessments across 844 tasks, found a consistent pattern: people want AI to take the low-value drudgery but prefer shared agency on high-value tasks, the machine executing while human judgment and responsibility stay in the loop. The same audit found investment flowing into automation zones workers explicitly do not want automated. The misalignment is not primarily technical. It is a management vacuum: nobody is deciding, on the record, which tasks move, which stay human, and what fills the freed hours.
That decision ledger is the manager's honest alternative to stealth. Task by task, not job by job: what the tool takes, what it drafts for human revision, what it must not touch, written where the team can see it. Freed capacity reassigned to named work, the backlog, the customers never called, the learning time, rather than left as invisible slack that finance eventually notices and harvests. Junior development deliberately rebuilt, since the old apprenticeship of routine tasks is exactly what the tool consumed; if entry-level work is automated, entry-level learning has to be designed on purpose, inside the augmented workflow.
And the headcount question asked upward, in plain terms: is the plan a smaller team? A manager cannot control the answer, but she can refuse to let the answer be implemented through her silence. If the plan is fewer people, people can plan, retrain, apply elsewhere with a runway. What corrodes teams is not shrinkage; it is discovering that the shrinkage was policy all along and that the person running their weekly meeting knew the shape of it and performed normalcy.
The manager did not choose the technology, the market, or the CFO's spreadsheet. She chooses the redistribution, and she chooses whether it happens out loud. That is a narrower power than she would like and a larger one than she is using. Stealth is not a strategy her leadership imposed on her. It is a default she is allowed to decline.
Sources
- World Economic Forum (2025). Future of Jobs Report 2025. Employer survey, 1,000+ companies, 55 economies: 77% plan workforce upskilling in response to AI while 41% plan headcount reductions as AI automates tasks. https://www.weforum.org/publications/the-future-of-jobs-report-2025/
- Brynjolfsson, E., Li, D. & Raymond, L. (2025). Generative AI at Work. Quarterly Journal of Economics 140(2). 5,172 support agents; ~15% average productivity gain, up to ~34% for novices, near zero for the most experienced. https://academic.oup.com/qje/article/140/2/889/7990658
- Dell'Acqua, F. et al. (2025). Navigating the Jagged Technological Frontier. Organization Science. 758 BCG consultants; GPT-4 users completed 12.2% more tasks ~25% faster with higher quality inside the frontier, but made more errors outside it. https://pubsonline.informs.org/doi/10.1287/orsc.2025.21838
- Brynjolfsson, E., Chandar, B. & Chen, R. (2025). Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of AI. Stanford Digital Economy Lab. Workers 22-25 in the most AI-exposed occupations show a ~16% relative employment decline, concentrated where AI automates rather than augments. https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/
- Shao, Y. et al. (2025). Future of Work with AI Agents (WORKBank). Stanford SALT Lab, arXiv 2506.06576. 1,500 workers, 844 tasks; workers want automation of low-value tasks but shared agency on high-value ones; investment is misaligned with those preferences. https://arxiv.org/abs/2506.06576