Adrian Dunkley says Caribbean AI job losses are only part of the problem

Caribbean employers can reduce the demand for labour without announcing mass redundancies. They can leave vacancies unfilled, buy fewer hours from contractors and expect the same team to produce more. That is why I believe the region should watch wages and entry-level hiring alongside job losses.

I am Adrian Dunkley, President of the Caribbean AI Association. My concern is how employers distribute the gains from AI, and who still gets an opportunity to learn. A count of dismissed workers misses changes in pay, available hours and the route into skilled work.

The research separates tasks from lost positions

The Caribbean AI Risk Management Council's CAIRME employment-scenario method, dated 31 October 2026, separates technical task coverage, feasible substitution, adoption and an employer's decision to reduce positions. Its current pathway produces gross five-year displacement scenarios of 3.1% to 6.0% of current employment across the 12 economies with comparable sector data. Potential creation is modelled separately at 1.7% to 2.7%.

These are conditional calculations, not recorded dismissals, a regional forecast or an unemployment rate. The model uses broad agriculture, industry and services employment shares, representative task profiles and selected assumptions about adoption and employer response. Four economies lack comparable sector baselines and remain unscored. A country range does not establish a population-weighted Caribbean total.

This distinction matters beyond our model. The International Labour Organization's May 2025 exposure study finds that one in four workers globally is in an occupation with some generative-AI exposure. Exposure describes tasks that could change; it does not mean one in four workers will lose a job. Managers still choose how to organise the resulting work.

A busy department may still stop hiring

Consider a hypothetical customer-service team already working at capacity. AI helps prepare replies and search records. The business could hire more staff and serve more customers. It could also keep headcount unchanged while increasing case targets, or decline to replace the next person who leaves. A full inbox does not tell us which choice management will make.

My concern is that organisations will interpret faster drafting as evidence that every employee can absorb more responsibility. The time saved on a standard reply may have little connection to the time needed for a disputed claim, a safeguarding concern or an unusual transaction. Those cases require attention, evidence and judgement. Their consequences do not shrink because the first draft arrives quickly.

“A company can keep every employee and still make work less secure,” is how I would put it. “Watch the vacancies that disappear, the hours people can sell and the responsibilities added to the same job.”

That is an argument for measuring the complete service. A Jamaican insurer, for example, could compare accepted outputs, rework, escalations and customer complaints before raising a processing target. This is a proposed test, not a report of an insurer's results. If faster preparation increases unresolved exceptions, the apparent saving needs to be recalculated.

Wages depend on who can bargain for the gain

AI could increase the output of a specialist who understands both the subject and the system. That person may gain bargaining power if the firm depends on their ability to detect errors, handle exceptions and improve the workflow. A junior worker whose routine output becomes easier to purchase may face pressure on rates or fewer paid assignments. Neither wage movement is inevitable.

Ownership, competition, customer demand, labour protections and access to training can change that outcome. A small business may use AI to reach customers it could never previously serve, creating work for people around it. Another may use the same technology principally to reduce its wage bill. Calling both “productivity” conceals an important difference for households.

Concentrated expertise introduces a further problem. If only a few people can challenge the system, their absence becomes an operating risk. An organisation can become dependent on a supplier and on the handful of staff who understand it at the same time.

Closing the first rung could weaken the profession

Graduates often develop judgement by doing supervised work whose immediate output is modest: checking a reconciliation, drafting a basic report or explaining a routine exception. If firms automate those tasks and remove junior posts, they need another credible route for developing expertise. Otherwise, they may later complain about a shortage of experienced people that their own hiring choices helped create.

This is a long-term mechanism I believe the region should test, rather than a measured Caribbean trend established by CAIRME. Prospective students may avoid a field when its first jobs look inaccessible. That could weaken the supply of people in functions the economy still needs. It could also make higher pay for a small specialist group coexist with weaker prospects for everyone trying to enter.

Track the opportunities that a redundancy count misses

For an employer, the useful next step is a monthly record of vacancies opened and cancelled, junior appointments, contractor hours, pay bands, workload and error correction. Compare equivalent roles and record changes in demand. Include paid training and redeployment so that a staffing reduction is not treated as the only possible response.

By 10 November, choose one team and compare 20 completed cases before and after its AI workflow changed. Record the work a person had to redo, not merely how quickly the model responded. Keep personal data out of the review summary and ask staff whether the revised target leaves time to investigate exceptions.

For public agencies and universities, the question is whether new entrants can still acquire the experience employers require. Pair vacancy and wage information with employer interviews and graduate follow-up. Use the CAIRME map to examine assumptions, then collect the local evidence needed to replace them. The region needs to know who benefits from additional output and who is losing a route into work.