Executives are racing to deploy AI agents and trim costs, then asking a question they should have asked first: where will new talent learn the craft if entry-level work disappears? The strongest early labor-market data shows the hit is landing on young workers in roles most exposed to AI. Using administrative payroll records, early-career employees aged 22 to 25 in the most AI-exposed occupations saw a relative employment decline of about 16 percent, while experienced workers in the same roles held steady and less-exposed fields grew (Brynjolfsson, Chandar and Chen, 2025). The adjustment is happening through headcount rather than pay, and concentrates where AI automates tasks rather than augments them (Brynjolfsson, Chandar and Chen, 2025).
The risk is not only social. It is operational. If the first rung vanishes, you do not just lose a cohort of future managers. You lose the engine that trains them.
The first rung is cracking
Employers expect wide reshaping of work, and the direction is clear. The World Economic Forum’s 2025 survey of over a thousand employers across 55 economies projects 22 percent of jobs will be disrupted by 2030, with 170 million new roles created and 92 million displaced, a net gain of 78 million (World Economic Forum, 2025). Almost 40 percent of skills are expected to change, and 63 percent of employers cite the skills gap as the top barrier to transformation. In response, 77 percent plan to upskill their workforce, while 41 percent plan to reduce headcount as AI automates tasks. The Forum estimates that out of every 100 workers, 59 will need reskilling or upskilling by 2030 (World Economic Forum, 2025). These are employer expectations, not measurements, yet they map onto what the payroll data already signals at the entry level.
This is the classic apprenticeship problem in modern form. For decades, juniors learned by doing the repetitive work seniors could supervise quickly. AI is best at exactly that work. The result is a paradox: automation raises the premium on human judgment, but it removes the low-stakes repetitions people need to build it.
Managers are not set up to teach
The obvious fix is to make management more developmental. The system, as it stands, is not ready. In a 2025 survey of nearly 10,000 business and HR leaders across 93 countries, Deloitte reports that managers spend almost 40 percent of their time on firefighting and administration, and only 13 percent on developing their people. Over a third of managers feel unprepared for the people side of the role (Deloitte, 2025). At the same time, the middle is thinning. US employers were advertising 42 percent fewer middle management roles at the end of 2024 than in spring 2022, and Gartner, cited by Deloitte, projects that by 2026 one in five organizations will use AI to flatten their structure (Deloitte, 2025). These are surveys and market indicators, but they point in one direction. There will be fewer rungs, and less slack time in which a stretched manager can teach.
If you cut the layers that once absorbed coaching, you must replace them with process. Without it, the work speeds up while learning stalls. That is how you get brittle teams that look efficient until a non-routine problem arrives.
From job ladders to work charts
One blueprint comes from firms that use AI agents at scale. Microsoft’s 2025 Work Trend Index labels these “Frontier Firms” and finds that 71 percent of their employees say the company is thriving, compared with 39 percent globally. The report proposes a shift from a static org chart to a dynamic “Work Chart” that allocates people and agents to outcomes rather than boxes. In the same research, 82 percent of leaders say they will use AI agents to expand workforce capacity in the next 12 to 18 months, and 46 percent are already automating entire workflows with agents (Microsoft WorkLab, 2025). This is a vendor report with commercial interests and is based on surveys and product telemetry, so treat it as a design hypothesis. It still offers two useful moves: make work observable at the task level, and make learning a visible output.
Practical steps follow from the evidence and these designs:
- Rebuild the first rung as projects, not positions. If employment adjusts most where AI automates tasks, then pair automation with deliberate practice in tasks where AI augments humans instead. Use a “work chart” to assign juniors to outcome slices that are safe, frequent, and reviewable, so they get the repetitions that automation would otherwise erase (Brynjolfsson, Chandar and Chen, 2025; Microsoft WorkLab, 2025).
- Put development time on the manager’s scorecard. If managers only spend 13 percent of their time developing people today, make that figure visible and move it. Free time by routing administrative work to agents, then measure managers on coaching and observed skill gains, not only delivery speed (Deloitte, 2025; Microsoft WorkLab, 2025).
- Plan reskilling like a balance sheet item. If 59 out of 100 workers will need reskilling or upskilling and 77 percent of employers already plan to invest, treat it as a core capacity build rather than a side program. Tie reskilling to the tasks on your work chart, not to generic course catalogs (World Economic Forum, 2025).
- Protect the on-ramp in headcount planning. The payroll evidence shows early-career workers in AI-exposed jobs are the first to be cut. Adjust hiring plans so that every team running major automation keeps a defined junior quota and a path to real work, even as roles change (Brynjolfsson, Chandar and Chen, 2025).
None of this argues for slowing down adoption. It argues for instrumenting it. A flattened company without an on-ramp is a short-term win and a long-term liability.
The leadership read
Boards should ask three questions before approving the next wave of automation. First, show us the work chart: which outcomes are owned by people, which by agents, and where do juniors get safe repetitions. Second, show us the manager time budget: what share of time is spent on development this quarter, and how will agents shift it next quarter. Third, show us reskilling as a pipeline: how many workers moved from automated tasks to augmented ones, and what cycle time did it take.
Leaders who answer with outputs, not anecdotes, will find that the supposed trade-off between efficiency and talent is largely a design problem. The early data tells us who bears the risk when we ignore it. Young workers in automated roles are the canaries. With a redesigned first rung, they can be the first to benefit instead.
Sources
- Brynjolfsson, Chandar and Chen (2025). Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of AI. Using ADP payroll data, the authors find a ~16% relative employment decline for 22–25-year-olds in AI-exposed jobs, with adjustment via employment and concentrated in automation-heavy roles. https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/
- World Economic Forum (2025). The Future of Jobs Report 2025. Employer survey projecting 22% of jobs disrupted by 2030, major skills shifts, and broad plans for upskilling alongside headcount reductions. https://www.weforum.org/publications/the-future-of-jobs-report-2025/
- Deloitte (2025). 2025 Global Human Capital Trends. Survey finds managers spend ~40% on firefighting and admin and only 13% on development, with fewer middle-management postings and a forecast of AI-driven flattening. https://www.deloitte.com/us/en/insights/topics/talent/human-capital-trends/2025.html
- Microsoft WorkLab (2025). Work Trend Index: The Year the Frontier Firm Is Born. Survey and telemetry report describing “Frontier Firms,” the shift to a dynamic Work Chart, and broad leader intent to deploy AI agents to automate workflows. https://www.microsoft.com/en-us/worklab/work-trend-index/2025-the-year-the-frontier-firm-is-born