What AI Is Doing to African & Caribbean Remote Work
For about five years, professionals across English speaking Africa and the Caribbean have had something their parents' generation could barely have imagined. Legitimate, well-paid remote work for firms in Toronto, New York, and London, performed from a home office in Bridgetown, Kingston, Accra, or Nairobi. The pandemic forced employers to accept distributed work. Bandwidth and payment rails caught up. Time zones that once seemed inconvenient turned into an asset, with Caribbean workers overlapping the full North American business day and African workers bridging US and European hours.
The window opened wider than I expected. It is now closing in some places and changing shape in others, and the cause is artificial intelligence. The question is whether Caribbean and African remote professionals end up among its casualties or among its most effective users.
What "replacement" actually looks like in practice
The popular framing is wrong in one important respect. LLMs do not reliably do anyone's job. They are probabilistic text engines that hallucinate, drift, fabricate citations, and fail silently in ways that are catastrophic any time being wrong matters. Which is most of the time. A human still has to check everything the model produces. Once you have to check everything, you have not eliminated the work. You have moved it.
What is actually happening in offices is more honest, and worse. AI makes you roughly twice as productive at your job, and your employer responds by hiring half as many people in your role. From the worker's point of view, the difference is academic. When Dario Amodei, the chief executive of Anthropic, predicts that 50 percent of entry-level white-collar jobs may be gone by 2030, he is not saying the work will be done by chatbots running unsupervised. He is saying it will be done by half as many supervised humans.
The "human steers, machine assists" framing the labs put in their press releases is not what shows up in real workplaces. What shows up is the machine setting the pace and the human running to keep up, demoted into someone whose job is to sign off on whatever the model produces fast enough to keep the queue moving. Junior associates review briefs the model drafted. Support agents edit replies the model generated. Radiologists confirm the model's reads. The work has not been automated. It has been intensified. One person now produces what two used to. The remaining person is more stressed and less skilled, because she no longer does the apprentice-level work that built judgment, and the gains go to whoever owns the firm rather than to the labour that produced them.
For the Caribbean or African remote worker, the implication is clarifying. If you are paid by the hour or by the seat as a member of someone else's queue, the productivity AI gives you will not be yours to keep. It will be captured upstream by the firm that bills you out, and your reward for working faster will be that fewer of your colleagues are needed.
Why the window opened, and how AI widened it at first
The first effect of generative AI on cross-border remote work was, oddly enough, to make it more accessible to people in the global South. A junior analyst in Lagos could produce a deliverable that read like the work of a senior associate in London. A bookkeeper in Bridgetown could draft client communication with the polish of a much larger firm. The accent barrier on video calls did not go away. But the cultural-codes gap in written output that used to hold back many talented non-Western professionals more or less closed.
For a stretch in 2024 and 2025, demand for offshore remote talent climbed. AI had lowered the cost of integrating that talent: onboarding documents, code review, written communication, time-zone coordination. Caribbean and African professionals were beneficiaries of that arbitrage.
Why remote knowledge work is unusually exposed to intensification
The same forces that lowered the barrier to entry are now compressing the work itself. The categories most exposed are the ones that made cross-border remote employment possible in the first place.
Think about what makes work amenable to remote completion. The deliverable can be produced and transmitted digitally. The work can be specified in writing. Results can be evaluated against well-defined criteria. The worker rarely needs to be physically present with a client, a piece of equipment, or a colleague. Every one of those features also makes the job easier to intensify with an AI co-pilot. The digital characteristics that allowed the work to be done from Barbados in the first place (clean briefs, written deliverables, reviewable artifacts) are excellent training data and excellent grounding context for a model producing the first pass.
So fully remote knowledge work sits among the categories most exposed to the dynamic above. This is not a comment on the quality of the workers. Many are demonstrably better than the in-office equivalents they were hired to replace. The conditions that make a job portable are the conditions that make it intensifiable. A plumber in Atlanta is not running an AI co-pilot. A remote junior copywriter in Toronto is, and so is a remote copywriter in Cape Town serving the same client.
The professionals most at risk are the ones whose pitch to G7 clients has been "I can do the same competent, middle-of-the-distribution work as a domestic worker, for less." That arbitrage was always going to be temporary.
Where it lands hardest, and how fast
The most exposed categories are routine writing and editing, first-draft research and synthesis, bookkeeping and basic finance, tier-one customer support, standard graphic and presentation work, junior software development scoped to small well-specified tasks, and templated legal and compliance work. These are the categories that powered the remote-work-from-the-Caribbean-and-Africa growth.
The unfamiliar feature is not the disruption itself, but its speed. Bharat Ramamurti, a former deputy director of the US National Economic Council, observes that the China shock unfolded over several years; this one could compress into two. Anthropic's enterprise agent revenue went from $9 billion at the end of 2025 to a $30 billion run rate within months. The Oxford economist Carl Benedikt Frey put it sharply: "Most economists will acknowledge that technological progress can cause adjustment problems in the short run. What is rarely noted is that the short run can be a lifetime."
AI as a force multiplier, for whoever captures the gains
The dynamic above flips for the professional who owns the relationship with the client directly. The intensification trap only catches you when the productivity flows upstream to a firm. If you are the firm, in the sense of being an independent practitioner billing for outcomes rather than hours, with your own brand and your own client list, the gains accrue to you instead.
A consultant in Bridgetown who has spent twenty years building judgment about capital campaigns, board governance, or family-business succession can now produce work at three or four times the previous throughput. The bottleneck in that kind of practice was never the quality of thinking. It was the time required to write proposals, draft donor communications, model scenarios, and produce the meeting-grade artifacts that turn judgment into deliverables. An experienced solo practitioner can now serve a client portfolio that would previously have needed a small team.
There is also a reach effect. A Caribbean professional whose practice was historically limited by the size of the local market can credibly serve clients in larger ones, because producing polished client-facing work is no longer the constraint. The constraint becomes what AI cannot easily produce. A track record. A network. Demonstrated judgment in a specific sector. The trust of past clients.
A warning for younger professionals
Anyone in the first five years of their career needs to read the next bit twice. Anthropic's researchers found that junior engineers who leaned heavily on AI coding agents did not complete tasks much faster, and understood their own work less when later quizzed on it. If you delegate the task you were supposed to learn from, you arrive at thirty without the skills you needed by twenty-eight. The leverage I described above only works for professionals who already have judgment. Those still building it cannot offload the hard parts to a model and expect to develop the way previous generations did.
A second risk: political backlash in the markets we serve
US polling shows 79 percent of voters worried the government has no plan to protect workers from AI, and 72 percent concerned that AI is depressing wages. Populist anger at AI in our client markets is unlikely to stay neatly aimed at AI. Cross-border services delivered by remote workers in lower-cost countries are a familiar populist target, and some of the energy now aimed at automation will eventually be aimed at offshore labour as well, fairly or not.
A macroeconomic note worth keeping in mind
The aggregate picture is genuinely strange. Productivity rises. Output rises. Unemployment also rises. Wages stay flat or fall, because labour has no mechanism to capture productivity gains, and has not had one for the better part of fifty years. Politicians in our client countries will spend the next several years arguing about whether the economy is "actually" doing well, and both sides will be telling the truth about a different slice of the same data. Plan your career on the assumption that the headline growth numbers and the labour market reality are pulling apart, and that the policy environment in the US, Canada, and UK is going to be more turbulent than at any point in our working lives.
What this implies for our community
Four suggestions, written as much for myself as for anyone else.
Do not compete on tasks where the AI is your peer. You will lose on price even if you win on quality, because your client will eventually run the comparison.
Invest in the kinds of value AI cannot yet replicate. Relationships. Named accountability. Regulated credentials. Sector-specific track record. The kind of judgment that comes from having been wrong in interesting ways and learning from it.
Use AI heavily in your own production, but think hard about who captures the productivity. A named consultant who owns a brand and bills for outcomes captures her own AI gains. A subcontractor billed as a body in someone else's pipeline does not. Move from the second arrangement to the first as fast as you can, even if your starting day rate is lower.
If you are early in your career, protect your own learning. Use AI for the output, but do the skill-building your seniors did by hand. The mid-career market in five years is going to be unforgiving toward anyone who short-cut their twenties.
The remote-work window from the anglophone Caribbean and Africa to North America and the UK is not closing. It is narrowing in one part of the labour market and widening in another. The competent, generic, replaceable middle is going to get squeezed from both sides. The top, where deep expertise meets AI leverage and the practitioner keeps her own gains, is more reachable than it has ever been. Where each of us ends up will mostly come down to which track we choose in the next two years, which on the testimony of the people building this technology is roughly all the time we have.
