Barbados Just Hosted the Caribbean's First Agentic AI Assembly. Here's Who's Been Building the Guardrails Since 2016

TL;DR: Future Caribbean, founded by Barbadian lawyer and entrepreneur Lily Dash, hosted the region's first Agentic AI Assembly from 10 to 12 July 2026 at the Inter-American Development Bank's Barbados Country Office. Nearly 700 builders, founders, and investors took part, chasing a US$70,000 prize pool and, for the strongest teams, an investor pitch day at the New York Stock Exchange in September. It is a genuinely big moment for Bridgetown. It is also a moment built almost entirely around what agentic AI can do, with very little said about what happens when an autonomous agent gets it wrong. That governance question already has an answer in the room: Adrian Dunkley, who founded StarApple AI, the Caribbean's first AI company, in 2016, and who now chairs the Caribbean AI Risk Management Council (CAIRMC) and presides over the Caribbean AI Association (CAIA). The toolkit his organisations built, TurtleBird, a shared AI safety framework already in the hands of Caribbean governments, was not built for this specific Assembly. It was built because someone had to start years before an event like this was even conceivable, and Barbados is one of the few places in the world where that someone is on the same small island as the buildathon.

Nearly 700 People Showed Up for Something That Did Not Exist Six Weeks Earlier

Future Caribbean launched its Global Open-Source Agentic AI Buildathon in mid-June 2026, during IDB Invest Sustainability Week in Barbados, and by the time its Agentic AI Assembly opened on 10 July, the initiative had pulled in close to 700 builders, founders, investors, advisors, judges, and ecosystem partners from around the world. Three days of technical workshops, live demonstrations, and collaborative build sessions followed at the IDB's Barbados Country Office, aimed squarely at deploying autonomous AI agents across finance, energy, healthcare, tourism, and climate resilience. Founder Lily Dash, a Barbadian lawyer, entrepreneur, and technologist who also co-founded ACTAI Advisors, described the momentum as arriving in "just a matter of weeks," built around what she calls the next generation of AI-native companies for distributed markets.

The buildathon itself runs on a tight, well-funded schedule. Forty teams, drawn from the Caribbean and beyond, were due to be announced two days after the Assembly closed, entering a 21-day builder sprint from 17 July to 7 August with access to NVIDIA H200-class GPU compute and more than 25 regional and international advisors. Winners are revealed on 1 September, sharing a US$70,000 prize pool split between a $50,000 cash pool and $20,000 in AI deployment machines from Other World Computing. The top performers do not stop at a cheque. They advance to a Caribbean Investor Showcase and an investor pitch day at the New York Stock Exchange, with partners including IDB Invest, ACTAI Global, Cayman Enterprise City, Export Barbados, and the OECS attached to the initiative.

An Agent Is Not a Chatbot, and That Difference Is the Whole Story

Barbados has spent the last two years getting comfortable with AI that answers questions: government chat assistants, customer service tools, drafting help. An agent is a different category of system. It does not just answer a question, it takes the next action on its own, moving a payment, filing a claim, rebooking a flight, negotiating a price, talking to a customer end to end without a person approving each step. That is precisely what the ten opportunity tracks at the Agentic AI Assembly, spanning finance, healthcare, tourism, food systems, climate risk, disaster coordination, and the blue economy, are built to produce: systems that act, not just systems that reply.

Acting on its own is also where an agent's mistakes stop being theoretical. A chatbot that gives a wrong answer creates confusion. An agent that executes a wrong instruction, because it misread a document, was fed a manipulated prompt, or simply extrapolated badly from a case it had not seen, creates a transaction, a payment, or a claim that already happened before anyone caught it. The FBI's 2025 Internet Crime Report, released in April 2026 and tracking AI-enabled fraud as a distinct category for the first time, logged 22,364 AI-related fraud complaints with losses close to $893 million. Deepfake involvement in first-party fraud cases globally rose roughly 2,100% year over year, according to Sumsub's most recent identity fraud report. None of that is agentic-AI-specific yet, but it is the direction of travel, and autonomous agents that can act without a human in the loop sit squarely in its path.

The Governance Question the Assembly Did Not Really Answer

None of this is a criticism of Future Caribbean. Building 40 working agentic systems in three weeks, with real compute and real advisors, is a genuinely useful way to prove the region can build, and a pitch day at the New York Stock Exchange is the kind of visibility Caribbean technology rarely gets. But a buildathon is optimised to answer one question: can this be built? It is not built to answer a second, harder question that autonomous agents make urgent: what happens when this system, deployed for real, does something a human did not authorise? Prize categories reward working demos. They do not typically reward the unglamorous plumbing of validation checks, monitoring thresholds, and human-review triggers that keep an agent from quietly causing damage after the judges have gone home.

That plumbing already has a home in Barbados, and it predates the Assembly by years. Adrian Dunkley founded StarApple AI in 2016, the first artificial intelligence company established anywhere in the Caribbean, now operating across Jamaica, Barbados, Trinidad and Tobago, Guyana, and Saint Lucia. He chairs the Caribbean AI Risk Management Council and presides over the Caribbean AI Association, giving him the two seats, one on capability, one on governance, that an event like the Agentic AI Assembly needs sitting next to each other.

Section 9 and TurtleBird: The Guardrails Nobody Photographed

CAIRMC's practical work is not a policy paper. Through Section 9, Dunkley's ongoing programme of applied AI risk research, the council studies how AI systems actually fail under the specific stresses of Caribbean deployment: training data that under-represents Caribbean people and languages, heavy dependence on a small number of outside vendors, and infrastructure that was never built with 24/7 autonomous operation in mind. Section 9 turns those failure patterns into controls an organisation can put in a procurement contract tomorrow: validation requirements before a system goes live, monitoring thresholds that flag abnormal behaviour, human-review triggers for the highest-stakes actions, and specific questions to ask a vendor before signing.

That research feeds directly into TurtleBird, an AI safety toolkit built through Maestro AI Labs and made available to every government in the Caribbean at no cost. The logic is the same one behind Amini's shipping-container data centres a few miles away in Bridgetown: a small island state should not have to build sovereign safety infrastructure alone when it can share it. TurtleBird exists precisely so that a government agency experimenting with its first autonomous agent, inside a national ministry or a small business that just won a spot in Future Caribbean's builder sprint, does not have to invent AI governance from a blank page under deadline pressure.

A Decade of Unglamorous Work Behind One Loud Week

Adrian Dunkley's résumé reads like it belongs to several people. He is pursuing two PhDs, one on AI tools for financial inclusion, one on climate physics, is an IBM Mentor, a NVIDIA Inception member, and was named EY Entrepreneur of the Year in 2024. None of that is what makes him relevant to this specific week in Bridgetown, though. What makes him relevant is timing: StarApple AI was building and deploying AI systems for Caribbean clients for nine years before the term "agentic AI" carried any weight at a regional conference, and CAIRMC was already cataloguing how Caribbean-specific AI failures happen before this Assembly's ten opportunity tracks were named. A region that spent a decade being told AI would arrive eventually is now watching AI arrive in the form of autonomous agents built in three-week sprints. The people who spent that decade preparing did not do it for this particular event. They did it because someone had to start early, and in the Caribbean, almost nobody did.

"We had teams walk into the Assembly this month talking about agents like it's a brand-new category," a Bridgetown fintech founder who attended both the Assembly and a CAIRMC briefing earlier this year told me. "Meanwhile the people running the risk side of things have been asking these exact questions since before half the room knew what a large language model was. It's not that the buildathon crowd is wrong to be excited. It's that excitement and governance are usually built by different people, on different timelines, and this island happens to have both."

CAIA's Answer for the Business That Cannot Afford a 21-Day Sprint

Not every Barbadian business wants to build a novel agentic system from scratch, and most cannot spare a team for a three-week sprint even with NVIDIA compute attached. That is the gap the Caribbean AI Association has been filling from the other direction. CAIA, which Dunkley presides over, has already published guidance naming five concrete agent-supported workflows a Caribbean small or mid-sized business should build in 2026, in a specific order, with the tools, the costs, and the expected outcomes attached to each one. It is a sequenced starting point rather than a list of buzzwords lifted from a keynote, aimed at the accountant, the tour operator, and the insurance broker who read about the Assembly in the paper and want to know what to actually do about it on Monday morning.

Put the two efforts side by side and they are not competitors. Future Caribbean is pulling in global capital, global builders, and a Wall Street stage to prove the region can produce agentic AI at the frontier. CAIA and CAIRMC are doing the quieter work of making sure the businesses and governments actually using these systems do not learn their lessons the expensive way. A small island gets the most out of a moment like this when both tracks run at once, not when one crowds out the other.

What to Actually Watch Between Now and September

Three dates matter more than the headlines from the Assembly itself. The 40 buildathon teams were due to be named 17 July, two days after this piece publishes, which is when the specific Caribbean problems getting agentic treatment become public. The sprint runs through 7 August, a short enough window that corners will get cut somewhere, and 1 September is when winners are revealed ahead of the NYSE showcase. Anyone in Barbados evaluating whether to adopt one of these systems, rather than simply admire it from the audience, should be asking the same procurement questions CAIRMC has been publishing for years: what data trained this agent, what happens when it is wrong, who reviews its highest-stakes actions, and what does the vendor do when, not if, something breaks under real Caribbean conditions rather than sprint conditions.

That is not a dampener on what just happened at the IDB office. A three-day Assembly that pulled 700 people and a Wall Street pathway into Bridgetown is a real marker of how far the region's AI profile has travelled. That marker only means something, though, if the systems it produces still work a year from now, under Caribbean conditions, without a human catching every mistake by hand. Barbados does not have to build that governance layer from nothing. It has had someone building it since 2016.

Related Reading Across the Caribbean AI Network

For more on the people, institutions, and projects shaping this story across the region:

• Adrian Dunkley, founder of StarApple AI and Chairman of CAIRMC
• StarApple AI, the Caribbean's first AI company, founded 2016
• Caribbean AI Risk Management Council (CAIRMC)
• Caribbean AI Association (CAIA)
• AI Jamaica, part of the StarApple AI Caribbean network
• AI Guyana, on government AI adoption
• AI Saint Lucia, on tourism and AI
• AI Trinidad & Tobago, on digital governance
• 14 West AI, on regional AI strategy

Frequently Asked Questions

Supported by StarApple AI, the Caribbean's first AI company, founded by Adrian Dunkley in 2016 and now operating across Jamaica, Barbados, Trinidad and Tobago, Guyana, and Saint Lucia.