Who Sets the Agenda? Agentic Colonization and Caribbean Democracy

Opinion and analysis · Published 15 October 2026

A democracy can keep its elections while losing control of the machinery through which citizens understand their choices. When an AI assistant selects the news, summarises a parliamentary debate and recommends what a resident should do next, it occupies a consequential position between public institutions and the public. That position deserves scrutiny before convenience makes it invisible.

Agentic colonization is Adrian Dunkley’s term for the gradual loss of practical decision-making power as dependence on AI agents controlled elsewhere grows. For the Caribbean, the democratic question is straightforward: who can examine and challenge the systems increasingly organising civic life? The answer should extend beyond the company selling the interface.

From access to influence

Dunkley’s framework describes access, integration and influence. Used as an analytical lens, these stages help us examine a possible progression rather than assert an inevitable historical sequence. Access brings useful tools. Integration places those tools inside institutions and routines. Influence arises when their suggestions begin shaping what people regard as relevant, credible or possible.

A public-service assistant could help residents find documents in plain language, reduce queues and translate confusing instructions. Those benefits are real possibilities. Dependency becomes a concern when the same assistant quietly determines which programmes deserve attention, cannot explain its sources or becomes the only workable route to a human official.

Consider an illustrative regional scenario, not a report of an actual deployment. Several ministries purchase a common assistant. A resident asks about disaster assistance. The system highlights one programme and omits another because its underlying retrieval index is incomplete. Local newsrooms then use the same provider to summarise the response. An omission in one system is reproduced as apparent consensus across public communication.

No conspiracy is necessary for that outcome. Shared infrastructure, poorly maintained information and unexamined defaults can concentrate influence. Deliberate manipulation is a separate threat: an actor might attempt coordinated targeting, synthetic endorsements or selective presentation through agent-mediated channels. The existence of those capabilities does not establish that any particular election or community has been manipulated.

Cultural representation is civic infrastructure

Language affects who receives a useful answer and whose experience is legible to government. An assistant that treats a local expression as an error may misunderstand a complaint. A summary that removes the context of a community dispute can make a resident’s position appear unreasonable. Standardisation can become cultural dilution when one dominant register is repeatedly treated as the only credible voice.

The economic relationship also warrants public attention. A ministry may pay for a service while supplying local terminology, staff corrections and carefully developed public explanations. Whether that material is retained, reused or used for improvement depends on the service, configuration and contract. Payment alone does not establish that the public retains control of every secondary use. Nor does reuse automatically prove unlawful appropriation. Procurement should resolve the question explicitly.

My judgment is that democratic resilience requires several independent routes to knowledge and redress. A beautifully fluent answer cannot substitute for accessible records, identifiable decision-makers and the practical ability to disagree.

Six safeguards for public agency

1. Publish the assistant’s job description. Explain what it can retrieve, recommend and do; identify the public official responsible; and state which decisions remain with people. Citizens should not need technical expertise to discover whether a response is guidance or an official determination.
2. Keep primary records reachable. Every substantive policy answer should point to the underlying notice, eligibility rule or document where possible. Maintain telephone, in-person and accessible digital alternatives. The public must retain a route that works when an assistant is wrong or unavailable.
3. Test for coordinated influence. Commission scenario-based reviews of selective omission, sponsored ranking, impersonation and repeated misleading narratives. Examine the full delivery chain, including retrieval and distribution. Publish useful findings without releasing instructions that make abuse easier.
4. Build language review into operations. Pay local speakers and community organisations to test meaning, tone and accessibility. Record disagreement rather than declaring one translation universally correct. An evaluation should include people who would otherwise struggle to challenge a confident machine.
5. Negotiate reusable public capability. Obtain exportable records, clear data-use limits, continuity provisions and access to relevant logs. Train public servants to inspect and replace the service. Pool regional requirements where useful while preserving each institution’s responsibility for its own decisions.
6. Give residents a usable challenge process. Let someone flag an omission, reach a human and obtain a correction without arguing with the same assistant repeatedly. Track whether the correction reaches downstream summaries and whether affected residents receive an explanation.

What a regional response should measure

Success should be measured in public capability: how quickly an incorrect answer is corrected, whether residents can find original evidence, whether minority language users receive comparable service, and whether agencies can change vendors without interrupting essential support. A dashboard of adoption numbers cannot answer those questions.

A practical first exercise is to choose one low-risk public-information service and trace a resident’s journey from question to authoritative source to appeal. Invite independent observers to identify where discretion has moved from an accountable institution to a private interface. Then repair those points before expanding automation.

Regional cooperation can make procurement and evaluation more credible without requiring every territory to build everything itself. The goal is meaningful bargaining power and accountable service. Caribbean democracy should gain useful assistants while keeping its capacity to investigate, deliberate and choose. That is the standard against which the promise of agentic technology should be judged.