A September 2026 study by Bharat Chandar of the Stanford Digital Economy Laboratory and Bouke Klein Teeselink of King’s College London examines 1.25 billion job postings and 154 million employment records across 41 countries. The researchers infer generative AI adoption from job advertisements and compare adopting firms with control firms. Their estimates indicate that senior employment rose while the junior share fell.

The employment effect is uneven by seniority

In a separate report, Bloomberg points to headline estimates of a 6.7% increase in senior employment over five years at AI-adopting companies and a 3% decline in junior employment, with the junior share down 1.9 percentage points. The paper itself describes modest overall employment growth and says the fall in junior share is driven mainly by senior growth rather than a broad collapse in junior headcount. The causal interpretation remains subject to the study’s design and assumptions.

Short term productivity can create a long term gap

Experienced employees are often better placed to use AI because they can judge sources, recognise exceptions and define a good result. Replacing some junior tasks may therefore improve near-term output. The risk appears later if organisations remove the work through which people learn the business, observe difficult cases and build professional judgement.

Insurance and global benefits are particularly exposed to this tension. Junior analysts learn by reconciling data, reading contracts, checking country rules, preparing renewals and discussing exceptions with experienced colleagues. Some of that work is repetitive, but it also reveals how information, wording and decisions connect.

Assistants should teach as well as complete

A well-designed assistant can preserve learning while reducing unproductive effort. It can show its sources, explain why a rule applies, flag uncertainty and ask the user to validate critical steps. Junior staff can compare their analysis with the system, while senior reviewers focus on material judgement rather than correcting formatting or searching for documents.

Training workflows can deliberately expose users to representative cases. The system can withhold the final answer until the analyst records an initial view, then provide feedback and route unusual cases to a mentor. This approach turns the assistant into part of the apprenticeship rather than a substitute for it.

Boards should monitor capability formation

Workforce dashboards usually measure headcount, cost and vacancies. They should also track the flow of junior hires, time to competence, review quality, promotion rates and the availability of supervisors. For AI-enabled teams, management should identify which tasks were removed and how the underlying skill is now acquired.

GBKxAI can make this a product feature. Its workflows can preserve reasoning steps, link answers to the corpus and provide a review trail that supports coaching. The commercial message is not that every manual task should remain. It is that productivity gains should be accompanied by an explicit plan for producing the next generation of experts.

Redesign entry level work deliberately

Managers should identify the junior tasks that are disappearing and classify the skills those tasks previously developed. Some skills can be learned through simulation, supervised case review or rotation. Others require exposure to live clients and exceptions. The workforce plan should then create replacement learning opportunities instead of assuming that employees will acquire judgement from the assistant itself.

This approach also clarifies the role of senior staff. Their time must include review, explanation and coaching, not only approval of difficult outputs. Organisations can set a target number of reviewed cases, structured feedback sessions and observed decisions for each development stage. The result should be assessed through accuracy, independence and progression. If junior hiring falls while the need for senior expertise rises, the talent model will eventually become unsustainable.

Benefits providers can apply the same principle to client education. An assistant that produces an answer without showing how local regulation, policy wording and data led to it may save time but add little capability. Explanations should be concise and linked to the decision. Over time, usage data can show where teams repeatedly rely on the system and where targeted training is needed.

Sources : SSRN – How Does AI Change Labor Demand Evidence from 41 Countries