The first years of a career used to be a trade: you did the grunt work and learned how the place ran. AI is taking the grunt work. Leaders say they intend to use digital agents to expand workforce capacity in the next year and a half, and nearly half already automate entire workflows with them (Microsoft WorkLab, 2025). That is good for speed. It is bad for apprenticeship if the tasks that taught novices are the first to go. Early employment data offer a warning. In occupations most exposed to AI, workers aged 22 to 25 saw a relative employment decline of about 16 percent compared with peers, while more experienced workers in the same occupations held steady and employment grew elsewhere (Brynjolfsson, Chandar and Chen, 2025). The authors find the labor-market adjustment is happening through hiring and headcount, not pay cuts, and that the risk concentrates where AI automates rather than augments.
The first rung is splintering
Automation is not new. What is new is the speed and scope of this wave, and the way it targets exactly the tasks that once served as practice. Employers expect that nearly two in five job skills will change by 2030, and they foresee widespread task reconfiguration as they deploy AI: 77 percent plan to upskill their workforces, 41 percent plan headcount reductions as tasks automate, and about half plan to reorient their businesses to new AI-driven opportunities (World Economic Forum, 2025). The Forum’s forecast is net positive on jobs by 2030, with 170 million new roles offsetting 92 million displaced for a net gain of 78 million, but it is also explicit that 22 percent of roles will be disrupted and that a skills gap is the top barrier leaders cite to business transformation (World Economic Forum, 2025). Those are employer expectations, not outcomes, yet they map to what the ADP payroll data already show for young workers in AI-exposed roles.
The mechanism matters. Brynjolfsson, Chandar and Chen separate tasks where AI substitutes for human effort from those where it amplifies it, and the employment damage shows up in the substitution case (2025). If firms hand full workflows to agents, the very tasks that taught new hires disappear. If instead they pair agents with people to speed research, structure analysis, or draft first versions, entry-level staff still touch the work and learn the craft.
Managers cannot absorb the training burden as is
Even if leaders want to protect apprenticeship, most managers lack the capacity. Managers report spending nearly 40 percent of their time on firefighting and administrative work, and only 13 percent on developing their people. More than a third say they feel unprepared for the people-leadership part of the job (Deloitte, 2025). The career ladder, meanwhile, is thinning in the middle. U.S. employers were posting 42 percent fewer middle-management openings at the end of 2024 than in spring 2022, and Gartner, cited by Deloitte, expects a fifth of organizations to use AI to flatten their structures by 2026 (Deloitte, 2025). Fewer rungs and overloaded managers is not a recipe for healthy pipelines.
AI can help here, but only with intent. The same agents leaders use to automate workflows can take over the low-value inbox triage, meeting scheduling, and reporting that clog a manager’s week. Microsoft’s Work Trend Index reports that 82 percent of leaders plan to use agents to expand capacity over the next 12 to 18 months (Microsoft WorkLab, 2025). If that freed capacity does not show up as time coaching juniors, it will be lost in yet more meetings and status updates.
Design a modern apprenticeship
Some firms are already experimenting with a different operating model. Microsoft describes “Frontier Firms,” companies that deploy AI agents across the enterprise and organize around a dynamic Work Chart focused on outcomes rather than fixed roles (Microsoft WorkLab, 2025). The label is marketing, and the evidence is a mix of survey and product telemetry, but the managerial design choices underneath are practical.
Three moves travel well across contexts:
-
Keep beginner work human in the right places. Start from the Brynjolfsson, Chandar and Chen distinction. Where AI substitutes cleanly, use it but pair it with shadowing and structured debriefs so juniors still see the steps. Where AI augments, route those tasks deliberately to early-career staff. Treat the agent as a power tool that speeds their hands, not a replacement that removes the job (2025).
-
Make coaching a measured output. Deloitte’s numbers show managers devote only 13 percent of their time to people development (2025). Set a hard target to raise that share and track it, and remove work to hit it. Give agents the routine admin that consumes almost 40 percent of a manager’s week, then hold leaders to account for observable development outcomes—documented skill gains, internal certifications, and progression rates for entry-level hires.
-
Redraw the work in teachable units. The Work Chart idea is useful here. Break outcomes into modules with explicit learning objectives, map which can be automated end-to-end and which should be human-led, and assign novices to the latter with agent scaffolding, templates, and checklists (Microsoft WorkLab, 2025). Use this map to target upskilling dollars where the World Economic Forum says leaders will spend them anyway, and to close the skills gap employers rate as the main blocker of transformation (World Economic Forum, 2025).
These are operating decisions, not HR slogans. They decide whether AI turns your entry-level cohort into observers of polished outputs or participants in the messy middle of real work. They also decide whether managers become “agent bosses,” a term Microsoft uses for leaders who design, supervise, and integrate agent workflows, or whether they become traffic cops for software nobody really owns (Microsoft WorkLab, 2025).
The practical read
The evidence is incomplete, but the signal is clear. The best administrative data we have, from ADP payroll records, shows that young workers in AI-exposed jobs are absorbing the early shocks, and that the shock comes through hiring and headcount rather than wages (Brynjolfsson, Chandar and Chen, 2025). Employers themselves expect sweeping task change, sizable displacement, and a decisive need for upskilling by 2030 (World Economic Forum, 2025). Managers as currently configured do not have the time or preparation to pick up the slack (Deloitte, 2025). And the leaders most bullish on AI are already automating full workflows, which risks hollowing out the first rung unless they redesign it (Microsoft WorkLab, 2025).
Treat apprenticeship as a design constraint in every AI deployment. Before you automate a workflow, decide how a newcomer will still learn it. Before you buy another agent, decide which hours of managerial admin it will erase and how those hours will fund coaching. The ladder will not break everywhere. It will break in firms that forget why the first rung existed in the first place.
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 show a roughly 16% relative employment decline for 22–25-year-olds in AI-exposed occupations and that impacts track automation rather than augmentation. 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 surveys project 22% of roles disrupted by 2030, a net gain of 78 million jobs, large skills shifts, and significant upskilling plans, framing the scale of redesign required. https://www.weforum.org/publications/the-future-of-jobs-report-2025/
- Deloitte (2025). 2025 Global Human Capital Trends. Finds managers spend nearly 40% of time on admin and only 13% on developing people, many feel unprepared, and mid-management postings have fallen, underscoring a capacity gap for apprenticeship. https://www.deloitte.com/us/en/insights/topics/talent/human-capital-trends/2025.html
- Microsoft WorkLab (2025). Work Trend Index 2025 — The Year the Frontier Firm Is Born. Reports that 82% of leaders plan to use AI agents to expand capacity, 46% automate workflows, and outlines the Work Chart and agent-boss concepts that inform operating designs. https://www.microsoft.com/en-us/worklab/work-trend-index/2025-the-year-the-frontier-firm-is-born