My estimate is that the Caribbean has about ten years to set its own terms for AI: who owns the data, who writes the rules, and who builds the systems that run inside our businesses and ministries. After that, the standards and platforms will be settled elsewhere and the region will be adapting to them. The gap is fixable, and the fixes are specific.
The number that worries me most comes from applications to a Caribbean startup grant fund. More than half of the founders applying for AI funding say AI integration is in their plans. Fewer than one in ten can describe, with any technical detail, how an AI agent would run inside their business.
That gap between intention and ability to execute is a fair measure of where the region sits today: closer to consumer than builder. I have watched the same pattern through mobile and cloud since I started building AI in Jamaica in 2011. We arrive well informed and a little late. With AI, late costs more, because the decisions being made now are about data ownership and infrastructure, and those are hard to reverse.
Tool use will spread by about 2028; the harder layer will not
On current trends, I expect everyday use of AI tools across Caribbean businesses to be common by roughly 2028. Prices keep falling, and the models handle Caribbean English and Spanish better every year. Tool use is not where the region is short.
The shortfall is in the layer underneath, which needs institutions rather than individual founders:
- governance rules written by people who know how AI systems fail, not only that they should be regulated;
- data ownership agreements, so Caribbean data producers see a return on what their data trains;
- training and test data that reflect how Caribbean economies work, including the informal economy;
- agent software configured for local business processes, currencies, and regulators;
- university programmes that keep AI talent working on Caribbean problems instead of exporting it.
Jamaica's National AI Task Force, where I serve as a member, is doing real work on this. So are UWI and a number of individual builders. The pace of that work is still slower than the pace at which global AI infrastructure is being deployed, mostly because the governance and data layers are underfunded.
AI Capacity Debt: ambition without the means to deliver it
I use the term AI Capacity Debt for the gap between a country's stated AI ambition and its institutional ability to deliver on it. A national AI strategy with no funded talent programme is a debt. A data protection law with no regulator able to assess an AI system is a debt. Most Caribbean jurisdictions carry some of both.
The debt is invisible at first, because nobody has tried to close it yet. It shows up years later, when a ministry wants to procure an AI system and finds that no one on staff can evaluate the vendor's claims, or that the data it needs sits on a platform whose terms it never negotiated.
Agents make the problem concrete. An agent built for the US market and deployed in Trinidad will handle the parts of a workflow that match US assumptions and get the rest wrong without raising an error: a VAT rate, a bank's file format, a parish name it does not recognise. Those failures do not announce themselves, which is why local testing matters more than the vendor's demo.
Why I put the window at ten years
The estimate rests on two trends and one generation.
- Agent standards are forming now. I expect the current agent platforms to reach a settled architecture in five to seven years. The APIs and integration norms agreed in that period will be hard to change later, and the region needs credible technical people in those rooms.
- Caribbean data is already being used. Financial, health, agricultural, and informal-economy data from the region feed global systems today with little return to Caribbean institutions. Once those extraction patterns are normal, governments need more political capital to reverse them than most more political capital than most governments will spend.
- The next cohort of professionals. People entering the workforce this decade will either learn to build and configure AI or learn only to use what others built. Only builders get a say in how the systems are designed.
This is a judgement, not a forecast with a confidence interval. The weakest part of the argument is the five-to-seven-year figure for agent standards: it could be shorter if one or two platforms dominate quickly, and longer if open standards keep fragmenting.
Four priorities, in order
- Sovereign data infrastructure. Registries, data standards, and institutional agreements that give Caribbean data producers ownership of what their data generates.
- AI-literate oversight. Regulators and procurement officers trained to test how agents behave and where they fail, not generalist committees.
- Production experience. Funded routes from university programmes into real deployments, so graduates have shipped a system before they leave, not only built a demo.
- Regional coordination. CARICOM has the mandate. It needs a regional AI action plan with named owners and quarterly reporting, rather than another statement of intent.
Most small economies face the same window. The Caribbean's advantages are real: firms used to building under tight budgets, dense informal networks that spread practice quickly, and regional solidarity when it is organised. Those advantages only count if they are pointed at infrastructure as well as products.
What to do next
- Founders applying for AI funding: before your next application, write one page describing a single agent in your business: its inputs, the systems it touches, what it outputs, and who checks it. If you cannot write that page, the application is not ready.
- Ministries and state bodies: for every AI procurement over your internal threshold, require the vendor to run a test on your own local data, with your staff present, before contract signature.
- Data protection regulators: name at least one staff member responsible for AI systems and fund their training this financial year, so that complaints about automated decisions have someone to go to.
- Universities: make a deployed project with a Caribbean business or agency a requirement for completing any AI or data science degree, starting with the next intake.
- Businesses buying imported agents: list the local specifics your workflow depends on (tax rates, bank formats, place names, regulatory filings) and test the agent against each one before it goes live.
Frequently Asked Questions
How can a small business test an imported AI agent against local conditions?
Build a test set of 20 to 50 real past cases from your own records, including the awkward ones: JMD and USD amounts, GCT, local bank statement formats, parish or district names, and customer messages in patois or Creole. Run the agent on them and compare its output with what your staff actually did. Record every miss before you decide whether to go live. Repeat the test whenever the vendor ships a major update.
Who regulates personal data used by AI systems in Jamaica?
The Office of the Information Commissioner administers Jamaica's Data Protection Act 2020, whose main obligations came into force on 1 December 2023. The Act applies to any AI system that processes personal data about people in Jamaica, including chatbots and agents run by overseas vendors on your behalf. As the business, you remain responsible as the data controller. Check the Commissioner's published guidance before deploying a system that makes or supports decisions about individuals.
Does the Caribbean have a regional AI governance framework?
There is no binding regional framework yet: no CARICOM-wide AI law with enforcement powers, so the rules that apply are each country's own data protection, consumer protection, and sector laws. Several territories have national AI task forces or strategies in progress, and governments have drawn on international references such as the OECD AI Principles and UNESCO's Recommendation on the Ethics of AI. A business operating across several islands should check each jurisdiction's data protection law separately.
What should a Caribbean government ask an AI vendor before signing a contract?
Ask where the data is stored and processed, whether it is used to train the vendor's models, and how you get it back if the contract ends. Ask for results from a test on your own local data, not a generic demo. Liability for a wrong decision about a citizen should be agreed up front and written into the contract. If the vendor cannot answer these in writing, treat that as the answer.
Do Caribbean countries need to build their own foundation models?
No. Training a frontier model costs far more than any Caribbean budget can justify, and the models available through commercial APIs already handle Caribbean English reasonably well. The better use of money is local evaluation data, fine-tuning or retrieval over Caribbean documents, and the skills to deploy and check these systems. Owning the data and the test sets gives more control than owning a model that falls behind within a year.
The infrastructure gets built somewhere
Ten years is less time than it sounds. Mobile and cloud caught most Caribbean governments and businesses off guard in the 2010s, and AI is moving faster. The region has builders who understand both the global tools and the local context. What they need now is public and private money directed at data, oversight, and production skills. If that layer is not built here in the next decade, it will be built elsewhere, on someone else's terms.