The Disappearing Ladder

Executive Summary

The talent pipeline that produces your future leaders is breaking, AI is automating the tasks that made up the bottom rungs of the corporate career ladder: research, first-draft analysis, data cleaning, basic code. The junior cohort isn’t being fired, it’s just not being hired, and the consequence of this is a slow-motion collapse of the system through which organizations develop the judgment, instinct, and institutional knowledge that senior leadership requires.

For CEOs, this will transform from a headcount problem into a succession problem.

The Entry-Level Contraction

A study from Stanford’s Digital Economy Lab shows that, among workers aged 22 to 25, employment in AI-exposed occupations roles declined 13% between late 2022 and late 2024, even as hiring for experienced workers in the same roles grew. 

This matters because it is one of the few studies that attempts to separate AI’s impact from the broader post-pandemic hiring correction, interest rate effects, and the tech sector’s own cyclical contraction. Most of the headline figures circulating (50% declines, 46% drops in graduate roles) conflate all these factors and attribute the result to AI. The truth is messier but no less concerning.

Even after controlling for noise, AI adoption is producing a measurable, specific compression of the entry level. Thirteen percent in roughly two years is not catastrophic, but the trend line matters more than the snapshot, and two features of AI adoption make linear extrapolation dangerous.

The first is velocity. AI capability is compounding, the tools available to a hiring manager in early 2025 are materially more capable than those available in late 2022, when the Stanford study period began. Tasks that required human judgment eighteen months ago (nuanced writing, multi-step analysis, code review with contextual awareness) are now within reach of widely available models. Every capability gain expands the set of entry-level tasks that can be automated. The 13% figure almost certainly understates where we are today, let alone where we will be in 2027.

The second is adoption speed. The Burning Glass Institute tracked experience requirements in AI-exposed fields between 2018 and 2024. In software development, the share of jobs requiring three years of experience or less fell from 43% to 28%. In consulting, from 41% to 26%. In data analysis, from 35% to 22%. The OECD’s 2025 Employment Outlook confirmed this pattern across member nations: firms are raising experience thresholds in cognitive roles because AI has reduced the need for the task execution that juniors used to provide.

These are not marginal shifts, they represent a structural recalibration of what companies expect from their workforce, and that recalibration is accelerating as AI tools become more embedded in daily operations.

It is worth being clear about what this is not, it is not a uniform collapse across all industries. Healthcare entry-level postings rose 13% over the same period, driven by aging populations and chronic staffing shortages. Trades and physical-presence roles remain robust. AI-adjacent roles (machine learning engineers, AI product managers) are growing rapidly. The contraction is concentrated in cognitive, desk-based functions where the daily work consists of tasks AI can now perform; drafting, data processing, research, basic analysis.

This picture is more precise than “all entry-level jobs are disappearing,” and in some ways more troubling since it means that the specific entry-level roles that served as the training ground for knowledge workers are being automated.

The Organizational Shape Is Changing

The traditional organizational geometry (wide base, narrow top, predictable ascent) assumed that routine cognitive work required human labor and that supervision was the mechanism through which both quality control and professional development occurred. AI breaks both assumptions simultaneously.

This is already visible. Revelio Labs reports a 40% drop in middle-management job postings since 2022, though the scope and sector breakdown of that figure deserve scrutiny. Gartner projects that by 2026, 20% of companies will use AI to flatten hierarchies, eliminating more than half of mid-tier management roles. Although it is a projection, not an empirical finding, it is consistent with what we are currently observing. 

If the base of the pyramid is compressing, the pyramid itself cannot hold its shape.

In our work, we see two competing shapes replacing the pyramid. The distinction between them is the difference between an organization that thrives in the AI era and one that will end up hollowing itself out.

The first is the obelisk or diamond. The base narrows: fewer entry-level hires, but those who join function as AI orchestrators from day one, managing and validating automated outputs. The middle widens: this becomes the largest layer, staffed by professionals focused on nuanced judgment, cross-functional integration, and contextual interpretation that AI cannot provide. The top sharpens maintaining institutional expertise and toward strategic AI integration and ethical governance.

The obelisk/diamond is the rising aspirational model for knowledge-intensive firms. It requires deliberate investment in the middle layer and structured pathways from entry to seniority. Without that investment, organizations drift toward the second shape.

The hourglass. In this model, automation targets mid-level cognitive tasks from above while eliminating entry-level tasks from below. The result is a bifurcated workforce with high-paid strategists at the top and low-paid AI support roles at the bottom. The middle, where professionals historically developed the judgment and relationships that qualified them for leadership, hollows out. An hourglass organization can function in the short term, but cannot reproduce and propagate itself, it has no succession bridge.

The difference between an obelisk and an hourglass structure is design and intent. Both are responses to the same AI pressures, the obelisk results from deliberate organizational investment, the hourglass results from cost-driven automation without redesign. CEOs are making this choice right now, whether they realize it or not.

What Remains Human

The traditional career deal (trade your labor on routine tasks for mentorship, learning, and eventual promotion) is being renegotiated. The tasks that made up “learning by doing” are the same tasks AI can now perform faster and, in a growing number of cases, better. This creates an uncomfortable question: if the training ground is automated, how do you develop the professional skills needed for a successful career?

We find the capabilities that matter most in the professional setting are the ones AI cannot replicate. MIT Sloan researchers Roberto Rigobon and Isabella Loaiza developed the EPOCH framework to identify where humans complement rather than compete with AI: Empathy, Presence, Opinion and Judgment, Creativity, and Hope and Leadership. What distinguishes EPOCH from the usual “soft skills” lists is empirical grounding: tasks requiring high EPOCH scores showed stronger employment growth between 2016 and 2024, and new roles emerging in 2024 carry significantly higher EPOCH scores than pre-existing ones.

To simple categorize these as “soft skills” would be incorrect, there is nothing soft about them. Teaching someone to solve a math problem is relatively straightforward, teaching them to exercise judgment under ambiguity, or to build trust with a skeptical client, is far harder and far, far more valuable. Rigobon’s research validates what practitioners have long observed: the human capabilities that matter most are the ones that are hardest to teach and impossible to automate.

Consider what this looks like in practice. An AI can draft a perfectly competent financial analysis, however it cannot sense that the CFO presenting the numbers is uneasy about a figure, probe the source of that discomfort, and surface a risk that the data alone wouldn’t reveal. That is empathy and presence combined with judgment in action.

An AI agent can generate twelve strategic options for market entry. It cannot sit in a room with a founding team, read the political dynamics between the CEO and the head of product, and recommend the option that is not only analytically sound but actually executable given the people involved. That is opinion, creativity, and leadership operating together. 

These are not edge cases, they are the daily substance of high-value professional work, and they require the kind of tacit knowledge that develops only through years of practice in complex human environments.

This is where the talent pipeline risk becomes existential. Tacit knowledge (the experience-based intuition that separates a competent analyst from a trusted advisor) has historically been acquired through exactly the kind of repetitive, structured early-career work that AI is now automating. Junior lawyers developed judgment by reviewing thousands of contracts. Junior consultants developed client instincts by sitting in hundreds of meetings taking notes and watching how senior partners navigated disagreement. Junior engineers developed systems thinking by debugging code they didn’t write.

Remove those experiences and you lose the mechanism through which expertise is transmitted across generations.

The emerging model is what some are calling the career lattice: non-linear, skill-led progression with lateral moves across functions, where competency rather than tenure determines advancement. In a lattice model, a three-year AI specialist may mentor a twenty-year veteran on tool adoption, while the veteran mentors the specialist on client management. Value is measured by skill density, not chronological seniority. This is conceptually appealing and practically difficult, because it requires companies to build skill-based compensation systems, internal mobility platforms, and deliberate apprenticeship programs that transfer tacit knowledge without depending on the routine-task pipeline that used to do it naturally.

What CEOs Should Do

The structural shift described above is not a five-year planning exercise. The velocity of AI capability improvement means that the organizations making deliberate choices in the next twelve to twenty-four months will set the terms for the next decade. The following is where we believe CEOs should focus.

Map your value chain against AI capability, honestly. Most organizations have a vague sense that AI is “changing things.” Few have done the granular work of mapping which tasks, in which roles, and at which seniority level, are already being performed faster and cheaper by available tools. When we’ve done this with clients, the results are consistently surprising: 30 to 40% of tasks in analyst and associate cohorts are already automatable with off-the-shelf tools. Neither the analysts nor their managers typically know this.

Start with your highest-volume knowledge-work functions. Baseline current performance metrics: cycle time, cost per output, error rates. Identify where AI agents can replace discrete task execution versus where human judgment remains essential. This audit is not a cost-cutting exercise, it is meant as the foundation to underscore every subsequent decision about organizational design.

Redesign the middle and the entry level. The temptation is to hire fewer juniors and pocket the savings. This works for two or three years, then you discover that your mid-level pipeline has dried up, your senior hires from outside don’t carry institutional knowledge, and your organization has lost the capacity to develop its own leaders.

The better move is to redesign what both layers mean. New joiners should function as AI orchestrators from day one, managing automated workflows, validating AI outputs, handling the exceptions and judgment calls that models cannot. Build intensive apprenticeship structures that transfer tacit knowledge deliberately, through structured mentorship and exposure to complex decision-making, rather than through the routine task repetition that AI has rendered obsolete.

The middle of your organization is either your greatest asset or your biggest vulnerability. In an obelisk model, mid-level professionals become the core: they interpret AI outputs, exercise contextual judgment, integrate across functions, and manage the human dynamics that determine whether a technically correct recommendation actually gets executed. This requires investment in skill development, in internal mobility platforms that enable lateral movement, and in compensation structures that reward capability rather than tenure. 

The alternative is the hourglass: a hollowed middle, a broken succession bridge, and an organization that can execute but cannot adapt.

Build for velocity, not for the current state. Whatever AI can do today, it will do materially better in eighteen months. Design your organizational structure and talent strategy with this acceleration in mind. The tasks that still require human judgment in 2026 may not require it in 2028. This does not mean panicking about automation, it means building the organizational discipline of continuously monitoring AI capability shifts in your domain and adjusting role definitions, skill requirements, and workflow design accordingly. Companies that set their AI strategy once and revisit it annually are already falling behind those that treat it as a continuous operating rhythm.

Preserve institutional knowledge before it walks out the door. If your experienced workers retire or leave before their tacit knowledge is captured and transmitted, no amount of AI investment will compensate. Document decision-making heuristics, structure mentorship that doesn’t depend on routine-task proximity, create forums where senior practitioners articulate the judgment calls that machines cannot replicate: the client instincts, the risk intuitions, the contextual awareness that took decades to develop.

This knowledge is your organization’s most irreplaceable asset, and it is the one most at risk in a rapid AI transition. The organizations that will come through this transition strongest are those that redesign their structures, talent pipelines, and development pathways around the reality that AI has permanently changed what constitutes entry-level work.

The obelisk or the hourglass. The choice is being made now.

Sources

1. Stanford Digital Economy Lab — Employment decline in AI-exposed occupations among workers aged 22–25 (2025). Via Fortune.

2. Burning Glass Institute — Experience requirement shifts in AI-exposed fields, 2018–2024. Via HR Executive.

3. OECD Employment Outlook 2025 — Cross-national analysis of AI’s impact on entry-level cognitive roles.

4. EU Joint Research Centre — AI usage among EU workers: 30% adoption rate, task-level breakdown (2025).

5. MIT Sloan — Rigobon & Loaiza, EPOCH Framework: human capabilities that complement AI (2025).

6. Revelio Labs — 40% decline in middle-management job postings since 2022. Via AllWork.Space.

7. Gartner — Projection: 20% of companies using AI to flatten hierarchies by 2026. Via AllWork.Space.

8. Anthropic Economic Index — AI usage patterns: 77% automation, 12% collaborative learning (2025). Via Fortune.

9. World Economic Forum — Future of Jobs Report 2025: employer workforce expectations.

10. Workplace Insights — London law firm partner on mid-level associate compression (2025).

11. MIT Sloan / Johnson & Johnson — AI-powered skills inference model for workforce gap analysis.

This analysis reflects publicly available research and Vela Advisory’s observations as of February 2026. AI capabilities and labour market conditions are changing rapidly; readers should verify current data before making strategic decisions.