The people analytics systems of organizations that have adopted AI over the past three years show, almost invariably, the same picture: productivity rising, execution times falling, output quality improving. The KPIs being measured improve. The KPIs not being measured deteriorate in silence.

The most relevant figure from Deloitte's Human Capital Trends 2024, surveying 14,000 global executives, is not the 76% who recognize AI as an amplifier of individual performance. It is the distance between that 76% and the 12% who have conducted a formal assessment of AI's impact on junior talent development pathways. Those 64 percentage points of distance do not measure an opinion. They measure a structural organizational blind spot.

What is happening in the zone no one is measuring

Organizations develop their experts and leaders through progressive exposure to cognitively demanding work: diagnosing real problems, making commercial decisions with verifiable consequences, conducting complex analyses under close supervision. In most cases this is not a deliberate mechanism. It is the formative byproduct of work executed directly, with one's own cognitive hands, under conditions of real consequence.

AI executes those tasks with increasing quality. The data documents this without ambiguity: Brynjolfsson, Li, and Raymond (MIT/Stanford, 2023) measure an average productivity increase of 14% with peaks of 35% for junior low-performers. Dell'Acqua et al. (Harvard Business School, 2023) document a 40% improvement in output quality on structured cognitive tasks.

The problem is not in these numbers. The problem is in what these numbers do not measure: the development trajectory of the professional who did not execute the task because they delegated it to the system.

The paradox of efficient delegation

Every manager who uses AI to complete in twenty minutes the analysis a junior would have spent three hours on is rationally optimizing their own time. Each individual decision is correct. The aggregate of those decisions produces an unintended effect: the progressive elimination of the work that was transforming inexperienced professionals into experts.

Vedrai Observatory defines this phenomenon as Future Capability Debt: the gap between the expertise an organization will need to possess in the future and what it is actively developing today. Like technical debt in software engineering, it accumulates silently, grows over time, and becomes structurally costly when detected late. With one relevant difference: technical debt has a name, is measured in technology governance processes, and generates conscious decisions to accumulate or reduce it. Future Capability Debt does not yet have a shared name in the HR governance processes of almost any organization.

The cognitive literature on expertise (Ericsson et al., 1993-2016) is unequivocal: high-level competence requires years of direct exposure to problems at the edge of the professional's current capability. Cognitively structured, well-defined tasks with measurable feedback are both the most suitable for automation and the most suitable for learning. This overlap is the heart of the problem.

Why current HR systems don't see it

Standard people analytics KPIs measure the present: productivity, engagement, turnover, individual performance. None of these indicators captures the development trajectory of junior cohorts over time. An organization can have excellent productivity metrics and a simultaneously weakening leadership pipeline, and standard reporting systems would not reveal the contradiction.

Brynjolfsson (Stanford HAI, 2024) introduces the concept of complementarity threshold: the point beyond which AI and the professional substitute rather than complement each other. In seniors with solid judgment, this threshold is high: their expertise guides, challenges, and contextualizes AI output. In juniors, the threshold is structurally lower because the cognitive structures needed to do that critical work are still forming. Without a deliberate learning strategy, juniors risk becoming validators of AI output before becoming producers of autonomous judgment.

Productivity without pipeline is not efficiency. It is a short-term loan against your organizational future, at an interest rate no workforce planning model is currently calculating.

The missing metric: measuring the trajectory, not the snapshot

Vedrai Observatory has developed the Leadership Pipeline Risk Index (LPRI), a tool designed to be integrated into annual talent review cycles without replacing existing systems. It comprises three dimensions: the share of cognitively demanding work still executed directly by junior professionals; the amount of AI expertise developed by seniors that is documented and transferred to the rest of the team; and the speed at which current junior cohorts develop autonomous judgment competence compared to pre-AI cohorts at the same tenure.

The third dimension carries double weight in the calculation because it is the only one that distinguishes accumulated competence from the capacity to manage AI output. It is measured through structured assessments at 12, 24, and 36 months on scenarios not resolvable with standard AI, compared against internal historical benchmarks. An LPRI deteriorating for two consecutive years signals structural debt in formation and requires a review of the learning strategy regardless of productivity results.

Three questions every CHRO should bring to the next board meeting

Are junior cohorts in our critical functions developing autonomous judgment competence at the same rate as pre-AI cohorts, at equivalent tenure? If we are not measuring this, we do not know. And not knowing is the condition in which debt accumulates fastest.

Do we have an explicit policy on what share of junior formative work must remain in direct execution, function by function? Or is this decision made by default by each manager individually, without a conscious organizational choice?

Do our succession plans incorporate the risk that candidates identified today have accumulated less direct formative exposure than the cohorts that preceded them at equivalent tenure? If succession timelines are calibrated on pre-AI trajectories, they are overestimating the available pipeline.

‍