Executives argue about office days while the first rung of the ladder is giving way. AI systems now do much of the work once handed to juniors, from drafting briefs to sifting data. The result is a quiet squeeze on entry-level employment and on-the-job learning. A large new working paper using payroll records shows where the pressure falls first: on the youngest employees in jobs most exposed to AI (Brynjolfsson, Chandar and Chen, 2025).
Where the damage lands
Using administrative data from ADP, Erik Brynjolfsson and co-authors find that workers aged 22 to 25 in AI-exposed occupations experienced a relative employment decline of about 16 percent, while experienced workers in the same occupations, and workers in less exposed fields, saw employment hold steady or rise (Brynjolfsson, Chandar and Chen, 2025). The adjustment is happening through headcount, not pay. Wages did not fall in tandem, suggesting firms are hiring fewer juniors rather than cutting everyone’s compensation. The losses concentrate in roles where AI tends to automate human tasks, not where it clearly augments them. That distinction matters for how firms should respond.
This is one study, albeit a large one, and the authors are careful about causality. Still, the pattern is consistent with what many managers see: automation narrows the pool of “starter tasks” that used to justify entry-level hiring. If the first rung disappears, the pipeline that feeds tomorrow’s specialists and managers starts to dry up.
Flattening without apprenticeships is a dead end
The career ladder was already wobbling. Deloitte reports that managers spend nearly 40 percent of their time on firefighting and administrative work and just 13 percent developing their people, and more than a third feel unprepared for the people side of the job (Deloitte, 2025). In the United States, employers were posting 42 percent fewer middle-management jobs at the end of 2024 than in spring 2022, and Gartner, cited by Deloitte, projects that by 2026, 20 percent of organizations will use AI to flatten their structures (Deloitte, 2025). Fewer rungs and less coaching time mean fewer chances to learn by doing.
Surveys of leaders signal more change ahead. The World Economic Forum projects that 22 percent of jobs will be disrupted by 2030, with 170 million roles created and 92 million displaced, a net gain of 78 million (World Economic Forum, 2025). Nearly 40 percent of skills are expected to change by 2030, and 63 percent of employers cite the skills gap as the top barrier to transformation. In response to AI, 77 percent plan to upskill workers and 41 percent plan to reduce headcount as tasks are automated. These are expectations, not outcomes, but they point to a future with more automation and a bigger training burden. The figure that should focus boards is this: out of every 100 workers worldwide, 59 will need reskilling or upskilling (World Economic Forum, 2025).
If firms flatten and automate without rebuilding how people learn, they will win a quarter or two of efficiency and then stall. No one will be ready for the roles AI cannot do.
The agent era can help or hurt
Microsoft’s 2025 Work Trend Index, a survey of 31,000 knowledge workers with Microsoft 365 telemetry, reports that 82 percent of leaders plan to use digital agents to expand workforce capacity in the next 12 to 18 months (Microsoft WorkLab, 2025). In companies that deploy AI organization-wide and actively use agents, the report says 71 percent of employees perceive the firm as thriving, compared with 39 percent globally. It also proposes replacing the static org chart with a dynamic “Work Chart” organized around outcomes, and notes that 46 percent of leaders already use agents to fully automate workflows or processes (Microsoft WorkLab, 2025). This is a vendor-backed view, and its telemetry reflects Microsoft’s tools, but it captures a real direction of travel.
The question is not whether agents can take work. They can. The question is whether leaders will use them to free up time for coaching and project-based learning, or to strip out entry roles and hope the market supplies mid-career talent later. The Stanford evidence suggests that when AI automates rather than augments, juniors exit and the ladder hollows (Brynjolfsson, Chandar and Chen, 2025). The pragmatic choice is to design for augmentation so that agents become practice fields for novices rather than substitutes for them.
How to rebuild the on-ramps
-
Make development a first-order metric. Managers already say they lack time and preparation to grow people (Deloitte, 2025). Use AI to remove administrative work, then measure managers on documented coaching hours, skills progress and internal mobility. If automation does not translate into more development, it is a missed investment.
-
Design learning-rich workflows. Start by mapping tasks where AI can draft, summarize or analyze, then assign entry-level employees to review, edit and explain the outputs. The Stanford paper’s core finding is that augmentation and automation have different labor effects (Brynjolfsson, Chandar and Chen, 2025). Treat that as a design constraint. Build review loops and shadowing into agent-driven processes so juniors practice judgment, not just button clicks.
-
Use dynamic project marketplaces. Microsoft’s “Work Chart” is a useful metaphor for internal marketplaces that match people to outcomes in real time (Microsoft WorkLab, 2025). Rotate early-career employees through short, scoped projects with clear deliverables and feedback. This recreates the breadth of exposure that entry roles once provided, without recreating layers of bureaucracy.
-
Protect cohort hiring, even if roles look different. The World Economic Forum expects large-scale disruption and a heavy reskilling load by 2030 (World Economic Forum, 2025). Carve out apprentice-style positions where the job is to learn by working alongside agents and experienced staff. Title them differently if you must, but fund them consistently.
-
Build agent governance as a teaching ground. As firms automate entire workflows, someone must set prompts, check outputs and manage exceptions. Make that a stepping-stone role for juniors. It aligns with the move to agents reported by Microsoft and turns oversight into applied training (Microsoft WorkLab, 2025).
-
Tie learning to mobility. If middle-management slots are fewer, define alternative pathways. Lateral moves across domains can compound skills faster than waiting for a title that may never open. Given that 63 percent of employers see the skills gap as their top barrier, creating visible, skill-based moves is a direct way to address it (World Economic Forum, 2025).
The practical implication is simple. The scarcest resource in the AI era is not models or cash. It is managers who can teach and systems that let novices practice. The evidence shows early-career workers in AI-exposed jobs are the first to be sidelined, and organizational layers that once supported apprenticeship are thinning (Brynjolfsson, Chandar and Chen, 2025; Deloitte, 2025). Companies that turn agents into training grounds, measure development like output and protect entry routes will compound capability while others chase mid-career hires that do not exist at scale. That is what a future-ready ladder looks like.
Sources
- Brynjolfsson, Chandar and Chen (2025). Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of AI. Early-career workers in AI-exposed jobs saw a ~16% relative employment decline, with adjustment via headcount not wages, guiding how firms should design augmentation. https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/
- Deloitte (2025). 2025 Global Human Capital Trends. Managers spend almost 40% of time on admin and just 13% on development, middle-management postings fell 42%, and AI-driven flattening is expected, underscoring the apprenticeship gap. https://www.deloitte.com/us/en/insights/topics/talent/human-capital-trends/2025.html
- World Economic Forum (2025). The Future of Jobs Report 2025. Projects 22% job disruption by 2030, a net gain of 78 million jobs, and that 59 of every 100 workers will need reskilling, framing the scale of the on-ramp challenge. https://www.weforum.org/publications/the-future-of-jobs-report-2025/
- Microsoft WorkLab (2025). Work Trend Index 2025 — The Year the Frontier Firm Is Born. Reports broad plans to deploy AI agents, higher perceived thriving in agent-heavy firms, and a shift to a dynamic Work Chart, informing how to structure learning around agents. https://www.microsoft.com/en-us/worklab/work-trend-index/2025-the-year-the-frontier-firm-is-born