Caribbean AI Jobs Are Growing. Regional Productivity Is Not. The 2026 ILO Numbers Explain the Gap

Caribbean AI job postings are climbing while regional labour productivity is not: the ILO's 2026 Employment and Social Trends report puts Caribbean productivity growth at negative 0.6% a year across 2015 to 2025, against a 1.7% global average. Hiring for a skill and raising an economy's output are different things, and the gap between them is where most of the region's AI policy work now needs to sit.

A ministry adds an AI unit. A bank runs a prompt-engineering workshop for its analysts. A tourism board buys a chatbot licence. Each of those is a real, countable AI job or AI-adjacent hire, and taken together they explain why AI is one of the fastest-growing categories of job posting across the region this year. None of it, on its own, explains why the International Labour Organization is reporting a decade of shrinking output per worker across the Caribbean at the same time.

TL;DR

• The ILO's 2026 report found Caribbean labour productivity fell 0.6% a year from 2015 to 2025, against 1.7% global growth over the same decade.
• Over 51% of workers across Latin America and the Caribbean have been informally employed since 2015; informality sits at 54.6% in Jamaica, 62.0% in Barbados, and above 91% in Haiti.
• A joint ILO-World Bank paper estimates 26% to 38% of regional jobs could be touched by generative AI, with 8% to 14% seeing productivity gains and only 2% to 5% at risk of full automation.
• Roughly 17 million jobs across the region sit in reach of a generative AI productivity gain but are held back by a digital access gap, split almost evenly between women and men.
• Hiring for AI skills and closing the productivity gap are not the same project, and confusing them is the fastest way to spend a training budget on the wrong half of the workforce.

The Decade the Numbers Cover

The ILO report did not measure a single bad year. It looked at ten, from 2015 to 2025, and found the Caribbean losing ground on output per worker for most of them while high-income economies gained 1.1% a year on average. Set against the global figure of 1.7%, the regional gap is not a rounding difference. It is a decade in which the same worker, on average, produced slightly less each year rather than slightly more, in a period when AI tools were becoming available to make the opposite true.

The report frames this as a paradox rather than a simple decline, and the framing matters. Unemployment across the region has continued to fall, which on its own would normally read as good news. The same report finds job quality stagnating underneath that falling unemployment figure, with workers moving into jobs that count as employment in the statistics without adding much to what the economy actually produces. A region can look like it is putting more people to work while producing less per person, and both of those things can be true in the same report.

Where Informal Employment Breaks the AI Hiring Story

More than 51% of workers across Latin America and the Caribbean have sat in informal employment every year since 2015, according to the same ILO report. The number is not evenly spread. Jamaica sits at 54.6%, Barbados at 62.0%, and Haiti above 91%. An AI job posting, a corporate training programme, or a government upskilling mandate is built for a payroll relationship: an employer with a headcount, a training budget, and a role description. Over half the region's workers are not in that relationship at all.

That is the first place the hiring headline and the productivity number stop lining up. A formal-sector employer can post ten AI roles, fill all ten, and genuinely raise its own output, while the informal majority around it, street vendors, agricultural workers, gig-based service providers, sees none of that investment and none of that gain. National productivity is an average across the whole workforce. AI hiring, as currently structured almost everywhere in the region, only reaches the formal half.

What Generative AI Actually Touches

A joint ILO-World Bank working paper, Buffer or Bottleneck? Employment Exposure to Generative AI and the Digital Divide in Latin America, put a number on how much of the regional labour market generative AI could plausibly reach: between 26% and 38% of jobs across Latin America and the Caribbean. Split further, the paper estimated 8% to 14% of jobs could see real productivity gains from generative AI, and a much smaller 2% to 5% face genuine risk of full automation.

Two things follow from that split. First, the automation fear that dominates a lot of public AI discussion applies to a narrow slice of the labour market, not the sweeping share the word "automation" tends to suggest in a headline. Second, and more relevant to the productivity question, the larger group, the 8% to 14% positioned for a genuine productivity gain, only realises that gain if the worker in the role has the tools and training to use generative AI in the first place. A job being exposed to AI is a description of potential. It is not a report of anything that has actually happened yet.

The Digital Divide Is Half the Story

The same paper puts a figure on the gap between potential and delivery: roughly 17 million jobs across the region sit within reach of a generative AI productivity gain but are held back by a lack of digital access, split close to evenly between around 7 million held by women and 10 million held by men. That is close to half of every job the paper identified as exposed to generative AI in a way that could raise output.

Access, in this context, is not abstract. It means a stable enough connection to run a tool during a working day, a device suited to more than checking messages, and enough baseline digital fluency to move from curiosity to competent use inside a reasonable training window. A region with connectivity gaps between its capital cities and its rural interiors, and between its higher-income and lower-income households, will always see this 17-million-job gap concentrated in specific places rather than spread evenly. Closing it is an infrastructure and access programme wearing an AI label, not primarily an AI training problem.

Why the Skills Gap Needs Local Institutions, Not Imported Ones

A training curriculum built for a Silicon Valley workforce, and simply translated for a Caribbean audience, tends to assume the connectivity, hardware and existing digital habits that the ILO figures say a large share of the region does not have. National AI communities that build their own training from the actual conditions on the ground have a structural advantage here. AI Jamaica, for instance, runs practitioner-level upskilling calibrated to what Jamaican learners actually have access to, rather than to a template written for a market with universal broadband. StarApple AI, the first AI company built in the Caribbean, has run a similar test at the leadership level, training Caribbean boards directly in AI literacy and governance and publishing measurable results rather than simply issuing a policy memo. In both cases, the lesson is the same: a training programme that starts from the region's actual access constraints closes more of the productivity gap than one imported wholesale and rebranded for a local audience.

Where the Gap Shows Up First

Two sectors make the pattern easiest to see. Tourism runs on a mix of large formal employers, hotel groups, cruise operators, and a much larger informal layer of tour operators, drivers, vendors and seasonal staff who never appear on a single company's payroll. A hotel chain can deploy an AI concierge and a dynamic pricing model and post a genuine productivity gain in its own accounts, while the informal tourism economy around it, the majority of the people actually earning a living from visitors, sees none of the tooling and none of the training.

Business process outsourcing tells a sharper version of the same story. Jamaica's roughly 50,000-strong BPO workforce sits almost entirely in formal employment, which should make it the sector best positioned to absorb AI tools quickly. It is also the sector most exposed to the automation share of the ILO-World Bank estimate, since call-handling and scripted support work are precisely the tasks generative AI performs well. Where informality shields the tourism sector's workers from AI-driven job loss at the cost of shielding them from AI-driven productivity gains too, BPO workers face the reverse trade: formal employment brings them within reach of both the upside and the risk, often in the same job description.

What This Looks Like Inside a Single Employer

Picture a mid-sized Caribbean bank that decides to get serious about AI in 2026. It hires two AI-literate analysts, licenses a document-summarisation tool for its compliance team, and runs a two-day workshop for its branch managers. Every one of those actions is a genuine, countable contribution to the AI job postings and AI adoption figures that show up in a regional survey. None of it touches the tellers, the call-centre staff working from a script, or the contractors who clean and secure the branch network, all of whom remain outside the training budget and outside the productivity conversation entirely.

Multiply that bank by every formal employer in the region running the same limited version of AI adoption, and the aggregate picture matches what the ILO reported: a rising count of AI-related roles sitting on top of a labour market whose average output per worker has not moved. The fix is not to hire fewer AI analysts. It is to stop treating a hire in the AI unit as evidence that the organisation's AI strategy is complete, when it has usually reached a fraction of the people who actually do the work.

What the Next Report Should Show

The IDB projects Latin America and the Caribbean growing 2.1% in 2026, a modest figure against global uncertainty and the structural headwinds the ILO report describes. Whether AI hiring meaningfully moves that number over the next few years will not be answered by counting job postings again next year. It will be answered by whether the roughly 17 million workers currently locked out of the productivity gain by access, rather than by aptitude, have that access a year from now, and by whether training investment starts reaching the informal majority rather than stopping at the edge of the formal payroll.

CAIA's position is that a region hiring for AI roles while productivity falls is not a contradiction that resolves itself. It is a measurement of where the investment has gone so far, mostly toward formal, urban, already-connected employers, and a fairly precise map of where it has not: the informal majority, the access-constrained 17 million, and the workers whose jobs the ILO paper says are exposed to a gain they cannot yet reach. Closing that gap is a slower, less photogenic project than announcing an AI hiring drive. It is also the only version of this that shows up in next year's productivity number instead of next year's job-posting count.

None of this is an argument against hiring AI-literate staff, running training workshops, or buying AI tooling. Each of those steps is necessary. The point is narrower: they are necessary and not sufficient, and treating them as sufficient is how a region ends up with a decade of ILO productivity data that contradicts its own AI hiring numbers. The next Caribbean AI report worth writing will not count how many AI roles were posted. It will measure how many of the 17 million access-constrained jobs got connectivity, a usable device, and training that started from where those workers actually stand, not from a template built for a market that already had all three.

Adrian Dunkley, founder of StarApple AI and President of the Caribbean AI Association, has argued for several years that the region's AI strategy has to be built for the labour market that actually exists here, not the one a template assumes. The ILO's 2026 numbers are the clearest evidence yet for why that argument holds.