The World Bank's October 2026 regional update puts Trinidad and Tobago at 0.44 on an AI preparedness scale where the OECD average is 0.69. It says the region's binding limit is not access to AI models. It is the ability of workers, firms and governments to put them to use. For a retailer or small manufacturer, that points to one move this fortnight: fix one messy process, then test whether AI helps it.
The main findings
- The World Bank published its Latin America and the Caribbean Economic Update, October 2026, titled Harnessing the AI Transition. Its data cut-off was 23 September 2026.
- Trinidad and Tobago scores 0.44 overall. Its two weakest pillars are digital infrastructure (0.38) and regulation and ethics (0.43), each about 0.32 or 0.33 below the OECD average.
- The report says 78.1 per cent of the regional workforce works in jobs where AI has to deal with the physical world, so a focus on screen jobs misses most workers.
- Between Monday 12 October and Monday 26 October 2026, time one weekly process by hand for five working days, then with AI plus a named checker for five, and compare minutes and errors.
What the World Bank report says
The Office of the Chief Economist for Latin America and the Caribbean published Harnessing the AI Transition this month. Chapter 2 is the part that matters here. Its argument is that the region does not need to build multibillion-dollar frontier models to capture most of the economic benefit, and that the “main constraint is absorptive capacity”. In plain words: can your people, your firm and your ministry use the tool well once they have it?
The report takes aim at what it calls “screen bias”. Most AI debate assumes the exposed worker sits at a laptop. The Bank counts it differently. In its taxonomy, 8.4 per cent of the regional workforce are high-skill cognitive workers who stand to gain from AI as “strategic co-pilots”, and 9.5 per cent are routine cognitive workers facing immediate exposure. The largest group at risk of the next wave is 26.8 per cent: manual and transactional workers in structured settings such as factory assemblers, who face physical automation as robot costs fall. Another 19.7 per cent work in unstructured settings such as street vending and domestic work.
The World Bank's release accompanying the report puts the median share of working-age adults using generative AI across the region at 17 per cent. That includes uses outside work and does not measure gains in business productivity.
Firms come next. The report describes “broad but shallow” adoption: companies use AI to draft text and find information rather than to change core processes. The main barrier, it finds, is a deficit in managerial capability and technical knowledge, with cost a smaller factor. Its prescription for governments is to act as an anchor client for local AI tools, fix data governance, and fine-tune open-source models to local language and institutions, rather than chase “AI sovereignty” by building frontier models from scratch.
The numbers behind the headline
Table 2.1 of the World Bank report reproduces the International Monetary Fund's AI Preparedness Index (2023 edition) with World Bank calculations. Trinidad and Tobago's overall score of 0.44 sits below Barbados (0.50) and Chile (0.59), and just above Jamaica (0.43), Guyana (0.42) and Saint Lucia (0.37). Throughout this article, "OECD average" means the benchmark the report uses: the unweighted average of 34 OECD economies, excluding Chile, Colombia, Costa Rica and Mexico, which the report counts with Latin America and the Caribbean. The pillar scores explain where the points go.
| Pillar | T&T | OECD average |
|---|---|---|
| Human capital and labour-market policies | 0.49 | 0.65 |
| Innovation and economic integration | 0.45 | 0.65 |
| Regulation and ethics | 0.43 | 0.75 |
| Digital infrastructure | 0.38 | 0.71 |
The index measures national conditions, not what any given shop does. Its 2023 data cannot reflect policy changes since then. A separate measure, the 2026 Global Index on Responsible AI, places Trinidad and Tobago 86th out of 135 countries and jurisdictions. Its AI policy score is 13.05 out of 100. These are different indices with different methods; their scores are not interchangeable.
The report estimates another year of contraction for Trinidad and Tobago. The same World Bank update lists real GDP growth for Trinidad and Tobago at -0.5 per cent in 2025 and an estimated -0.2 per cent in 2026. For a small firm, that makes costs and returns worth checking carefully.
On the learning side, the report notes that only 19 per cent of young people in Latin America and the Caribbean complete secondary education with basic proficiency in reading and mathematics. The regional picture for ECLAC's tracking is broader: its Latin American Artificial Intelligence Index 2025 covers 19 countries across three dimensions: enabling factors, research, development and adoption, and governance.
Automation Fragility: fix the process first
The method library on this network describes Automation Fragility as AI amplifying existing weaknesses when it is deployed without fixing the underlying data quality and process gaps. The World Bank reaches the same place from the economics side: “AI cannot substitute for weak data foundations”.
Consider a shop with an unreliable stock count and an out-of-date supplier price list. An AI reorder suggestion can carry those errors into the next order. It can also introduce errors of its own. Check the stock records and price list first, then compare the tool's suggestions with the records a staff member has verified.
The test in the steps below is one we suggest, built on the Automation Fragility idea. It is our own application, not a published step of the method.
What this means in Trinidad and Tobago
The report gives no Trinidad and Tobago breakdown of these worker groups or local survey of how shops and factories use AI. The shares above are regional averages. Retail cashiers and factory assemblers appear in its examples of structured, physical work, but we cannot use those examples to count exposed jobs here. A local employer needs to examine actual tasks with the people doing them.
The readiness scores identify broad national gaps. They do not measure the broadband service at a particular shop or tell a business what legal remedies it has. Before buying a tool, test it on the connection and equipment staff will use. Ask the supplier about data storage, access and deletion, and check those answers against the contract. The World Bank also argues that hosting data centres should bring wider local benefits, such as supplier opportunities and skills. Construction and operations may create jobs too; broader gains depend on the project and its links with the local economy.
For a hardware shop on the Eastern Main Road, a useful starting point might be a weekly reorder list that staff spend too long correcting. This is a hypothetical example, not a reported local trial. The World Bank discusses “small AI”: focused applications designed for specific problems and low-resource settings. A national model or a large data centre is not a prerequisite for testing a narrow task.
For other Caribbean coverage, visit AI Barbados and AI Jamaica. Our small business guide has starting points at home.
How AI helps here
One study in the report offers a caution about promised time savings. In a randomised trial in Lima, Peru, teachers given a general-purpose AI assistant plus training used AI more, with the share naming it their most-used tool rising by 29 percentage points. They saved no detectable time. They did write better prompts and judged AI answers more accurately. Skill grew before speed did.
A teacher trial cannot establish what will happen in a shop. It does show why adoption and time saved should be measured separately. Drafting supplier messages, transcribing stock sheets and summarising sales are possible trial tasks. Count the time spent checking and correcting the output, along with subscription and setup costs. A faster first draft may still leave the whole task taking longer.
The risks: who could be hurt, and how
The report identifies manual and transactional workers in structured settings as exposed to physical automation: 26.8 per cent of the regional workforce. It describes categories of exposure, not a forecast that all these jobs will disappear. Employers should involve workers in trials and plan training before changing roles.
Data handling presents a separate risk. Customer lists, staff records and confidential supplier terms should not be pasted into an unapproved service. Start with synthetic or suitably de-identified examples, and check whether the business has permission to use any real records with the chosen provider. A named staff member should verify every output before it affects an order, payment or customer. An index score cannot establish the legal position; get appropriate advice where an actual data or contract issue arises.
What to do this week
- Pick one weekly process by Tuesday 13 October: the reorder list, the supplier invoice match, the production schedule or the delivery roster. Write down who does it, how many minutes it takes and what goes wrong.
- Measure the baseline: from Monday 12 to Friday 16 October, log minutes spent and every error found for that process, done as you do it today.
- Test it: from Monday 19 to Friday 23 October, draft the same process with an AI tool and have a named person check every line before it is used. Log minutes and errors again, separating errors the checker caught from any that got through.
- Review the results on Monday 26 October 2026. A 25 per cent reduction in total working time, with no error reaching a customer or supplier, could justify a longer supervised trial. Keep human checks in place. If the result is poor, examine the records, instructions, tool and review process before deciding what to change.
Ten working days and one process are too small a sample to establish reliability. Changes in workload can affect the comparison. Record the number and difficulty of cases, include review time, and treat any apparent saving as provisional.
Where this argument is weakest
The T&T score comes from the 2023 edition of an IMF index that measures a country, not a shop, and the worker shares are regional. The sources reviewed here do not provide a current Trinidad and Tobago firm-level AI adoption rate. The 25 per cent time target in the test is our own choice, not a benchmark from the report.
Questions readers ask
What did the Peru teacher trial in the World Bank report actually show?
In a randomised trial in Lima, a programme pairing a general-purpose AI assistant with training raised the share of teachers naming it their most-used AI tool by 29 percentage points. The teachers saved no detectable time on lesson planning or grading. They did write better prompts and judged AI answers more accurately, so the gain was skill rather than speed.
Where does Trinidad and Tobago rank on the Global Index on Responsible AI?
The 2026 GIRAI country profile places Trinidad and Tobago 86th out of 135 countries and jurisdictions. Enabling conditions score 56.54 out of 100, AI policy 13.05, and ethics and sustainability 20.33. The index cautions that indicator definitions changed between editions, so comparisons need care.
Would more data centres solve the readiness gap the World Bank describes?
Not by themselves. The World Bank argues that infrastructure investment should also support local suppliers, skills and businesses. Hosting capacity does not show whether a shop has reliable records, trained staff or a useful application. Those questions need their own evidence.
What does the World Bank mean by 'small AI'?
Small AI means low-cost, practical applications, small in model size or narrow in scope, that solve specific problems in low-resource settings and can reach millions through basic phones. The report borrows the term from the World Development Report 2026. A shop's reorder list or an invoice checker fits that description far better than a national language model.
Why does the report say workers in street vending and domestic work are not safe from automation?
The report groups 19.7 per cent of the regional workforce in routine work in unstructured settings, including street vending and domestic cleaning. It argues that low wages relative to robot costs can delay automation. These are categories of exposure, not predictions of which jobs will disappear or when.
Where can a small business owner get started without paying anything?
StarApple AI publishes the free LAC AI Playbook for micro and small businesses, in English, Spanish, Haitian Creole, Papiamento and Portuguese, at starapple.typeform.com/aiplaybook. Use it to choose a task for the supervised trial above. Two weeks can reveal problems worth investigating, but cannot establish that a tool is reliable.
Sources: the World Bank's official October 2026 report, including Table 2.1, Box 2.1 and the workforce analysis in Chapter 2; the GIRAI 2026 country profile; and ECLAC's ILIA 2025 summary. The preparedness index is the IMF's, as reproduced by the World Bank. Written for AI Trinidad & Tobago, supported by StarApple AI, which describes itself as the Caribbean's first AI company and was founded by Adrian Dunkley. Related: Caribbean AI Risk Management Council.