The entry-level work that nobody does anymore
In every large organisation, the trajectory of a graduate hire used to follow a remarkably similar pattern. The newcomer was assigned the work that nobody else wanted: compiling background briefings, cross-checking procurement data against policy frameworks, reconciling budget lines, drafting the first version of reports that a senior colleague would later reshape. The work was tedious, detail-heavy, and rarely acknowledged. But it served a purpose that no job description ever made explicit: it trained the eye. By spending months buried in the raw material of institutional decision-making, the junior learned to sense when numbers did not cohere, when a supplier's terms concealed a risk, when a recommendation rested on an assumption that nobody had verified. That instinct could not be taught in a classroom. It could only be grown in the field.
In the older tradition of structured apprenticeship, this process had a plain name: transmission. The apprentice did not begin with the masterwork but with the humblest gestures, and it was in that humility that mastery was forged. The modern corporation operated on the same principle for decades. Consulting firms, international organisations, and financial institutions recruited graduates in cohorts and put them through high volumes of foundational work, not because it was cheap labour, but because that volume was the crucible in which professional judgement took shape.
When AI absorbs the training ground of junior staff
The scenario is already unfolding. A senior programme officer can now feed a project brief into an AI agent and receive, within minutes, a structured situation analysis complete with risk matrices, stakeholder maps, and draft recommendations ready for circulation. For the experienced professional, this is an extraordinary gain: strategic oversight is preserved while the documentary production that consumed entire days compresses into a fraction of the time. In consulting and financial services, this shift is well advanced. In international programme management, procurement, and compliance, it is arriving fast.
The day corporate departments and international secretariats cross this threshold entirely, the graduate who joins those teams will find nothing left on which to train. The background research, the first drafts of policy briefs, the data reconciliation, the preliminary compliance checks — everything that constituted the junior's learning ground will have been absorbed by the machine. The new hire will arrive in an environment where the volume work, the very work that was supposed to teach them the profession, no longer exists.
This phenomenon deserves a name. Call it learning debt: the liability that organisations accumulate, without noticing, when AI replaces the work of their junior staff. Output increases; turnaround times shrink; apparent costs fall. But the source that once fed the next generation with competence, discernment, and the capacity for autonomous decision-making runs dry.
The freeze that is coming
Large-scale job cuts are not the most likely outcome. Organisations carry inertia, contractual obligations, and HR cycles that slow abrupt change. The most probable scenario is also the quietest: they simply stop hiring.
If, in 2026 and 2027, multinational corporations and international bodies scale back their graduate programmes because an agent handles volume work at near-zero marginal cost, the question answers itself: who will be the programme directors in 2032? Who will lead the strategic units in 2037? If the trend holds for two or three consecutive cycles, the fabric of mid-level competence tears apart. On one side stand capable tools; on the other, overburdened senior leaders with no one between them and the machine to ensure continuity. Unlike financial debt, which can be repaid in instalments, learning debt is irreversible. You cannot recover ten years of experience that a human being never lived through.
The trap of subsidised pricing
There is a dimension of this problem that decision-makers consistently underestimate. The current price of AI does not reflect its real cost.
The construction industry learned this lesson the hard way. Vendors of BIM software offered free licences to engineering and architecture schools, so that entire generations of graduates entered the market knowing only one tool. Design offices followed, out of necessity. Once dependency was established, the vendors eliminated perpetual licences, imposed mandatory subscriptions, and removed seat-sharing options. In 2020, more than two hundred European firms, among them Zaha Hadid Architects and Grimshaw, co-signed an open letter denouncing a seventy per cent increase in cost of ownership over five years. The firms that had built their entire production around those tools had no choice but to pay.
The pattern extends well beyond construction. ERP and CRM vendors followed the same arc: aggressive discounts during adoption, then steady price increases once migration costs made switching prohibitive.
Artificial intelligence is tracing exactly the same trajectory. Providers sell their services at a loss today, subsidised by billions in venture capital, to install dependency. The day those subsidies end, organisations that replaced their juniors with agents will face the worst of both worlds: teams with no pipeline of emerging talent, and tools that have become too expensive to remain viable. They will not have reduced costs. They will have mortgaged their future operational capacity.
Training differently: the junior as auditor of the AI
The diagnosis is sobering, but it does not condemn anyone. The solution is not to refuse AI; the solution is to redefine what it means to be junior.
The junior of tomorrow will no longer produce the first draft of an analysis or a report. The agent will handle that. Instead, the junior will need to learn how to audit what the agent produces: verifying its assumptions, identifying its blind spots, challenging its conclusions.
The new training ground is supervision and contradiction. Reviewing an AI-generated procurement analysis for a Geneva-based international organisation demands that the junior understand regulatory frameworks, institutional rules, and the political context in which a recommendation will land. It is an intellectually more demanding exercise than compiling a spreadsheet from scratch, and it develops professional intuition at a considerably faster pace.
Seniors, for their part, will need to accept a new role. Training a junior in an AI-augmented world means dismantling, in real time, what the machine has produced. It means showing where the answer is too smooth, pointing to the place where the agent missed a regulatory constraint, explaining why a solution that looks elegant rests on a fabricated data point. This is exactly what the master craftsman of old used to do: not executing in the apprentice's place, but teaching the apprentice to see what the untrained eye cannot distinguish.
Who will train the decision-makers of tomorrow when there is no one left to draft the first report?
Every organisation should be asking this question now, not in five years. AI has already replaced junior staff on volume tasks, and nothing will reverse that trend. Those who understand this will recognise that a junior was never a cost centre; a junior was an investment in the continuity of institutional knowledge. Eliminating that investment does not save money. It creates a debt. Silent, invisible, and irreversible.