Adrian Dunkley

AI Boss. Applied Physicist turned Founder. Building AI Empires before it was cool. 15+ years in A.I. Lover of Scifi & Storytelling.

[Adrian Dunkley @decisionmaestro]

Thank you to everyone helping with the Huricane Melissa relief efforts.
StarApple AI has donated $300,000 so far and will continue to support Jamaican.
Our teams are volunteering, travelling to affected areas, and providing food, equipment, and clothing. There are still families in danger and communities in need.
Please consider donating; any amount helps a family rebuild, find safety, and start again.
The clouds remain, but hope is rising. Let’s keep showing up for one another.
https://t.co/0fDYkG1lhn

#SupportJamaica #HurricaneMelissa #JamaicaStrong #Relief #CommunityFirst #Jamaica

[Adrian Dunkley @decisionmaestro]

61 FOR, 39 AGAINST

https://t.co/UUsPuNuk8t

#aiagents #bobmarley

[Adrian Dunkley @decisionmaestro]

Jamaican AI company to the Moon.

https://t.co/ZlusW1PWb8

#artemisii #AI #moonmision #caribbean #AI

[Adrian Dunkley @decisionmaestro]

Jogger Safety Mini App Prompt template:

Use the uploaded aerial image of my jogging trail to create a personal safety visibility analysis. The goal: Help me understand how sunlight, shadows, & darkness affect my trail at different times of day and year, so I can spot blind areas, poorly lit sections, and places that may feel unsafe while jogging.
Prioritize accuracy. Confirm the image is correctly oriented with north facing up, if not ask for a new one. My general location is <Country> <City or Region>
Use that information to calculate the sun’s movement accurately.
Once orientation and location are confirmed, analyze the aerial image, estimate how visible features such as trees, buildings, walls, or cliffs cast shadows, and simulate how light and shadow move across the trail throughout the day and year. Let me scroll through any time of day, including morning, midday, evening, and night, as well as different times of year.
Highlight blind spots, shaded areas, rapid visibility changes, and dark sections that may be unsafe, especially in early morning, sunset, or night conditions. Also flag areas that may need artificial lighting.
The final output should give a clear, interactive view of how visibility changes across the trail so I can plan safer jogging routes.

[Adrian Dunkley @decisionmaestro]

Personal Safety mini app in @AnthropicAI Claude in 2 minutes. (prompt in the comments)
Personal safety should be free and accessible. AI can help.

I started jogging, still negotiating with my lungs. During a run this week other joggers were talking about blind spots along the trail, areas that become darker earlier than expected, and sections that feel uncomfortable during evening and early morning runs. That did not occur to me, I've been running in the middle of the day.
A stretch of path that suddenly falls into shadow, a turn where visibility drops, a section where trees block the last of the evening light.
This is exactly the kind of everyday problem AI can help with. Then I saw one of the Claude demos, simulating lighting conditions around buildings.
So, I ran a quick Friday test using Claude’s new interactive feature. The idea:
1) Upload an aerial image of your trail,
2) confirm the location and orientation,
3) then simulate how sunlight and shadows move throughout the day and across different seasons.
The system can highlight sections that become dark earlier, identify potential blind spots, and show how visibility changes from early morning to sunset.
Something that used to require specialized modelling tools can now be explored in minutes. And Urban planners can use it too.

Anyone can analyze a jogging route, park, or hiking trail and better understand when certain areas become risky due to low visibility. That kind of insight should not be limited to urban planners or engineers. Personal safety should be free and accessible.
AI can help make that happen.
#AI #PersonalSafety #AIForGood #UrbanTech #Innovation #aisafety #ClaudeOpus

[Adrian Dunkley @decisionmaestro]

New Caribbean AI Startup Launched!

The Intelligence Layer powering AI built in the Caribbean and Latin American. The road to IPO begins.

https://t.co/ZlusW1PWb8

#aistartup #ai #caribbean #latam #agents

[Adrian Dunkley @decisionmaestro]

Most companies don't have a Strategy for AI; I don't mean an AI Strategy, I mean a Strategy for the why, and plan for the how and when of AI.
They have an AI subscription.
Those are not the same thing.
One is a tool. The other is a clear direction and a plan for how the tool changes what you do, who you hire, and what you stop doing altogether.
2027 is gonna be rough for those businesses.

[Adrian Dunkley @decisionmaestro]

OpenAI passed $25B in annualized revenue this week.
Alibaba's Qwen3.5 runs on high-end consumer hardware and is free to deploy.
DeepSeek V4 is approaching.
What does it mean for emerging economies?
The monopoly on frontier AI is cracking faster than Western coverage suggests. You no longer need a hyperscaler relationship to build serious AI infrastructure.
The window to build regional AI capacity is open. It will not stay open. Once the global platforms establish distribution dominance in your market, the technical gap becomes irrelevant.
Preparation Asymmetry is not permanent. But it rewards speed.

[Adrian Dunkley @decisionmaestro]

The Caribbean AI Playbook episode 1

https://t.co/8b5w3l5lVG

[Adrian Dunkley @decisionmaestro]

T&T created the Caribbean's first Minister of Artificial Intelligence. Only five countries had done it before them: UAE, UK, France, Canada. Then Trinidad.
That is either historic. Or a very expensive press release. The next three years will decide which.

What the ministry actually built in year one:
Anansi. A national AI assistant pulling services from 32 government agencies into one interface. A national AI readiness assessment with UNESCO and UNDP. Membership in the 50 in 5 digital infrastructure initiative.
For a first year, that is an innovation push!

Now the hard question nobody is asking publicly:

What does Anansi run on?
What happens when a citizen asks about their pension entitlement and Anansi is wrong?
A national AI assistant without published error rates and citizen correction mechanisms is not a public service. It is a liability waiting to be discovered.
Mottley said at Carifesta 2025 that the new colonisation would not come by sea. It would come through your phone.

T&T's minister disagreed with her framing. Both of them are partially right.
But neither speech answered the actual question: What is T&T building that the region can own?

Anansi is named after the spider trickster. Smart. Adaptable. Outmanoeuvres bigger opponents through wit.
The name is right. The strategy needs to match it.
Build something. Do not just deploy something.

Full analysis
https://t.co/BJfXmWLynj

#TrinidadAndTobago #CaribbeanAI #ArtificialIntelligence #AIGovernance #DigitalCaribbean #Anansi #CaribTech #MaestroAILabs

[Adrian Dunkley @decisionmaestro]

7.2 million visitor arrivals in 2023. The spending patterns, preferences, and behaviour of seven million tourists per year.
Barbados is sitting on one of the richest AI training datasets in the Caribbean. It is currently treated as a compliance record.
That decision is worth billions if it changes.
Full piece at
https://t.co/pErsfADlWj

#Barbados #MiaMottley #CaribbeanAI #AIPolicy #CaribTech #ArtificialIntelligence #DigitalBarbados #MaestroAILabs

[Adrian Dunkley @decisionmaestro]

Jamaica's AI Reckoning. A prioritization problem as old as the flag. The window to shape how AI touches Jamaican lives is open right now.
It will not stay open because we wrote a good document. Read the full breakdown at

https://t.co/Sk4eienBnw

#ai #caricom #jamaica #JamaicaTech #CaribbeanAI #AIPolicy #Jamaica #ArtificialIntelligence #AIRisk #CaribTech #MaestroAILabs

[Adrian Dunkley @decisionmaestro]

Is your Data Training Military Drones?

https://t.co/OPSAN8gyhH

#aiagents #war #AI

[Adrian Dunkley @decisionmaestro]

https://t.co/DUAiJvzVz7

[Adrian Dunkley @decisionmaestro]

Free AI Readiness and Training for Governments and Non-profits.
Taxpayers pay for wasted AI Spend, you can stop it.

Sign up link in the comments.

https://t.co/bynmcDnBuI

#aitraining #aigov #ainonprofits

[Adrian Dunkley @decisionmaestro]

The agent is ready. It can read your contracts, send your client emails, process your refunds, update your CRM, schedule your appointments, and file your reports. It will work at 2 AM on a public holiday without complaint. It will almost never forget a step in the process.

The process, though. That is your job to have sorted before any agent touches it.

Consider a hypothetical that plays out with uncomfortable regularity across Caribbean businesses: a financial services firm in Port of Spain deploys an AI agent to handle client onboarding queries. The agent is configured, tested, and launched. Within six weeks, it has given incorrect KYC guidance to a segment of applicants because the Central Bank of Trinidad and Tobago updated its digital onboarding requirements two months prior, and no one updated the document the agent was reading from. The agent did not fail. It executed perfectly against stale information. The firm carries the regulatory exposure either way.

That is the gap nobody talks about when they are selling you the platform.

Complete article below

https://t.co/DNEjlXHY2B

#aiagents #AI

[Adrian Dunkley @decisionmaestro]

4. Administrative and Public Sector Measures
Motor Vehicle Duty Concession Reform: The 20% duty concession regime for public officials has been modified. While the 20% customs duty rate and SCT exemption remain, the GCT exemption has been removed, meaning public officials must now pay GCT on these vehicle imports.
Projected Revenue: $1.3 billion (Effective May 1, 2026).

NHT Transfer: The government will continue the annual drawdown of funds from the National Housing Trust (NHT) to the central government for another five years.
Projected Revenue: $11.4 billion annually.

5. Future Implementation (Tourism)
GCT on Tourism Activities: The government announced an increase in the GCT rate for tourism activities (hotels, attractions, etc.) from 10% to the standard rate of 15%. However, to allow the sector to recover from Hurricane Melissa, this measure is deferred until April 1, 2027.
Continuing Relief Measures

Despite these new taxes, the government is maintaining the phased increase of the Personal Income Tax (PIT) threshold. For the 2026/27 fiscal year, the tax-free threshold will increase from $1.79 million to $1.90 million effective April 1, 2026. Additionally, to improve administrative efficiency, the filing deadline for Corporate Income Tax has been permanently shifted from March 15 to April 15.

#jamaica #taxes #melissa

[Adrian Dunkley @decisionmaestro]

🧵The Government of Jamaica has introduced new revenue measures designed to raise ~JMD 29.4 billion. These measures mark the first time in ten years that the Govt has introduced net new taxes, a shift they say are necessitated by the fiscal gap created by Hurricane Melissa (Cat5), which caused damage estimated at 41% of GDP in October 2025.

The major new tax measures introduced for 2026/2027 are categorized below:

1. Health and Social Taxes ("Sin Taxes")
The largest portion of the new revenue is derived from taxes aimed at public health and consumption behaviors.

SCT on Non-Alcoholic Sweetened Beverages: A new Special Consumption Tax (SCT) will be levied on non-alcoholic sweetened beverages (locally manufactured or imported) at a rate of $0.02 per milliliter.
Impact: A 300ml beverage will attract $6.00 in tax; a 2-litre beverage will attract $40.00.
Projected Revenue: $10.1 billion (Effective Q1 2026/27).

SCT on Alcoholic Beverages: The specific SCT rate on alcoholic beverages will increase from $1,230 to $1,400 per litre of pure alcohol (LPA).Projected Revenue: $1.6 billion (Effective May 1, 2026).

SCT on Cigarettes: The specific SCT on cigarettes will increase by $3.00, moving from $17.00 to $20.00 per stick.
Projected Revenue: $1.1 billion (Effective May 1, 2026).

[Adrian Dunkley @decisionmaestro]

2. Environmental Taxes
Environmental Protection Levy (EPL): The EPL rate will increase from 0.5% to 0.8% on both domestic and imported goods. Additionally, the tax base for locally manufactured goods will expand from 75% to 100% of the sales value.
Projected Revenue: $3.639 billion (Effective May 1, 2026).

3. Digital Economy Taxation
GCT on Digital Services: The General Consumption Tax (GCT) will now apply to imported digital services (e.g., streaming services, cloud computing) consumed in Jamaica. This measure aims to capture revenue from overseas providers who currently operate without taxing local consumption.
Projected Revenue: $300 million in 2026/27 (Effective Q4 2026/27).

[Adrian Dunkley @decisionmaestro]

https://t.co/xB7IZXGiOm

[Adrian Dunkley @decisionmaestro]

70 countries have published national AI strategies. 12 new ones in 2024 alone (Oxford Insights). More than half came from lower-income countries.
Good news. Also the start of a problem nobody is naming precisely.
Publishing a strategy is a statement of intent. It is not capacity. And the distance between what a nation promises in its AI strategy and what its institutions can deliver is where real damage accumulates.
I call this AI Capacity Debt.
AI Capacity Debt is the accumulating gap between a government's AI commitments and the institutional infrastructure required to deliver on them. Every strategy published without funded training adds to the balance. Every regulation drafted without regulators who understand machine learning adds to it. Every national AI lab opened without local datasets adds to it.
Like financial debt, it compounds. And the longer it goes unpaid, the more expensive the correction.
It accumulates in three stages. Stage 1: publish the strategy, get good press, appear on readiness indices. Stage 2: form the task force, produce recommendations, draft regulation. Stage 3: deliver. Retrain workers. Build data systems. Equip regulators.
Countries stall at Stage 3. Because Stages 1 and 2 are visible and cheap. Stage 3 is expensive, slow, and invisible until something breaks.
The compounding is what makes it dangerous. AI capability moves quarterly. Curriculum redesign moves in years. Job displacement follows the client's timeline, not the country's. And announced strategies reduce political urgency, because the visible work creates a feeling of completion while the delivery gap widens quietly underneath.
Where is AI Capacity Debt most acute? Countries dependent on a single AI-vulnerable sector. Jamaica's BPO employs 62,000 people. The Philippines' BPO employs 1.3 million. Countries with ambitious strategies but thin R&D bases. Jamaica: 0.06% of GDP in R&D, 13 AI publications in five years (UNESCO, 2025). And countries where trained AI workers emigrate to higher-paying markets. The investment is made. The debt payment is due. The asset has left the balance sheet.
UNDP's December 2025 report warned that AI may reverse decades of narrowing inequality between countries. AI Capacity Debt is the mechanism through which that happens at the national level.
The countries that will manage AI well are not the ones with the best documents. They are the ones that close the gap between what they announce and what they can deliver before the disruption arrives.
That gap is AI Capacity Debt. For most countries, the balance is growing.
Full article in the comments
#AICapacityDebt #AIGovernance #AIPolicy