A letter from Cebu, 24 October 2026. About fourteen minutes.

The wrong side of the AI map

Every guide to artificial intelligence assumes you live somewhere with fibre, a salary in dollars and a free afternoon. The countries those guides come from are having a difficult autumn. I live in Cebu, I lived in Antananarivo before that, and I run an American company from both. Here is what I have learned about using the machines without letting them use you, and an invitation to learn it with me.

By Wajdi.
Written in Cebu.

“Anything can happen in life, especially nothing.”

Michel Houellebecq, Platform

In Antananarivo the electricity went off most afternoons. You learn to plan a working day around it the way farmers plan around rain: whatever needs the grid gets done before two, whatever needs only a head and a notebook gets done after. From a flat in that city I ran a small American company. Its clients are on other continents, its work is made mostly of words, and the words have to be right the first time.

The work is reading and writing, mostly, which happen to be the two things the new machines do fastest and, left to themselves, worst.

I used them every day in Madagascar and I use them every day here. They draft. They sort. They turn four hundred pages of anything into a paragraph while the kettle boils. They also flatter, invent, and agree with whatever you said last. If you want to know what a year of that does to a person, you can read Eric So at MIT, or you can watch yourself.

The countries that write the manuals have had a difficult month

On Saturday 17 October, while I was drafting this, about 550,000 people marched in French cities against the cost of living and the state of their schools. By nightfall it was looting and fires in Paris, Marseille, Nantes and elsewhere; the town hall in Limoges burned. The Interior Ministry’s final count was 2,397 arrests in a single day. Curfews followed in parts of Paris, Bordeaux and Nice. On the Monday the state’s ten-year borrowing cost reached 4.93 percent; on the Wednesday it passed 5 percent for the first time since 2002, and the gap with Germany was the widest in fifteen years. Two of the three big rating agencies had already cut France’s grade. The third was reviewing it as I wrote.

The prime minister has no majority. His 2027 budget asks for 54 billion euros of savings to bring a deficit of 5.4 percent of GDP down to 5, the lower house has already stripped out some of the savings, and the decisive vote is on 3 November. In 2025 and again in 2026 the state financed itself through emergency laws because no budget could pass. Unemployment is 8.3 percent, the highest since 2020. The president, who cannot run again, had the approval of 17 percent of voters in mid-October, before the riots. Across the Rhine the chancellor has been sliding since spring and the AfD posted its best results yet in two state elections in September. Spain votes early, on 29 November. Serbia votes on 25 October after a march of 150,000 in Belgrade.

Dominique de Villepin, who was prime minister twenty years ago and wants to be president next spring, wrote on 18 October that the crisis had given way to chaos and that the French had the understandable feeling of no longer being governed. Three days later he described a country oscillating between stupor, shame and anger, and by the end of the week he was warning of a mechanism of civil war. Philippe de Villiers, from the other end of French politics, said on television that the collapse of the political class was above all an intellectual collapse, that France was no longer powerful because it obeyed a Brussels oligarchy that took its money, and, in the line I have not been able to shake, that problems prolonged without being solved become insoluble.

I quote them not because I agree with either but because they agree with each other, from opposite ends, that the centre has stopped holding. That matters on this side of the map for a plain reason. For sixty years francophone Africa and Madagascar have taken their cues from Paris: the curricula, the exams, in fourteen countries the currency itself. Nobody in Paris is drafting the manual for using these tools in Antananarivo this autumn. Nobody there is drafting it for Cebu or Kuala Lumpur either, but Cebu and Kuala Lumpur never expected them to. The Philippines and Malaysia have their own politics, which I will leave alone. What they share with Tana is the question of work, and I will come to it.

Everything about these tools is written somewhere else

The advice assumes fibre. It assumes the premium tier is a question of convenience rather than a question of rent. It assumes English, usually American English, and when it turns to French it produces the French of an insurance brochure. It assumes an afternoon free to experiment, a laptop that holds a charge, and a colleague down the hall who has already made the mistakes.

When I lived in Antananarivo my question was never which model was the smartest. It was which one would load before the battery died. The whole conversation about artificial intelligence has a map, and the map stops at the edge of the places where it is drawn. Madagascar is not on it. Neither is Mindanao, or Kelantan, or Casamance. That is not a complaint. It is a gap, and gaps are where small companies live.

The narrator of Houellebecq’s Platform, standing in Cuba in front of machinery he cannot repair, decides that he is perfectly adapted to the information age, that is to say, good for nothing. The novel is from 2001. People called it a provocation then. This month it reads like a weather report.

Three sets of numbers, none of them hypothetical

In France, 21.6 percent of people under 25 who want work cannot find it, against 15.5 across the European Union. The students who blockaded their lycées this month have read that figure. It is the one subject on which nobody accuses them of skipping the homework.

In the Philippines the headline rate jumped to 6 percent in July, 3.14 million people, and in April 7.4 million more had a job but wanted more hours than they could get. Cebu, where I am writing, lives on contact centres: around 1.9 million jobs nationally. In July the industry’s own association cut its forecasts for 2028, and its downside case, 1.85 million jobs, is fewer than exist today. Malaysia’s headline rate is 2.9 percent and its youth rate is 10.1; the distance between those two numbers is the whole story. In Madagascar unemployment is officially around 2 percent, which tells you only that 95 percent of work is informal and nobody can afford to be unemployed.

Then the number that connects them. Stanford’s Digital Economy Lab, working from the payroll records of millions of American workers, found that since late 2022 the employment of 22-to-25-year-olds in the occupations most exposed to AI, customer service and software development above all, has fallen by roughly 13 percent relative to everyone else, while older workers in the same jobs were untouched. The fall was concentrated where AI automates a task rather than helps a person do it.

The same tool, two ways of using it, and only one of them on the right side of that number. That is the whole case for learning this properly, and soon.

What Eric So found

Eric So is the Sloan Distinguished Professor of Global Economics and Behavioral Science at MIT, and this month W. W. Norton published his book, The Collision: What AI Does to Us. He studies behaviour, not chips, which is why his conclusion is uncomfortable: it is about us, not the machines.

He calls the problem AI gravity. Three forces pull in the same direction. A brain that wants to save energy. Classmates and colleagues who have already given in. Schools and employers that reward the giving in. Left alone, the pull wins, and the result is what he calls cognitive atrophy: the capacity to push through a hard problem, to remember what you read last week, to notice when you are being told what you want to hear, all of it thinning from disuse.

So sorts people into two groups. Those who use AI to think better, and those who use it instead of thinking. The tool is identical in both cases. What differs is who set the terms of the conversation: you, or the engagement team. It is the Stanford split again, this time measured in brains rather than jobs.

The comparison he reaches for is physical. A muscle that is asked for nothing weakens. I train four or five times a week, so I take that one personally. Nobody lost their legs to the lift. They lost them to taking it every time.

Among the studies So points to, the one that matters most to anyone reading this from Africa or Southeast Asia does not come from Boston. In 2024 the World Bank ran a six-week after-school programme in Edo State, Nigeria, in which first-year senior secondary students worked on their English with Microsoft Copilot. The prompts were designed with care. A teacher was in the room. The students put in about thirteen hours in total and gained roughly 0.3 standard deviations, which the researchers translate as somewhere between one and a half and two years of ordinary schooling. The same machine that is hollowing out term papers in Massachusetts. A different design.

That paragraph is the whole course, if you want the short version. The long version, from a man with tenure, is in a summary of So’s findings that ran this month in Nacionale.

Too much, all of it true, none of it useful

“The sheer volume of information dissolves the information.”

Günter Grass

Grass said that long before any of this existed. If you are reading this in Manila or Kuala Lumpur, in Dakar or Tana, you know the volume. Every third video is a young man in a rented sports car explaining ChatGPT. Every group chat has a PDF of five hundred prompts. Every course promises an income and a certificate.

I paid for the advertisement that brought you here. I am not above the system; I am inside it, and I would rather say so than pretend this is journalism. The difference I can offer is small and specific. I will tell you exactly what I am building before I ask you for anything, and you can read me for free for as long as you like before you decide whether I am worth paying.

Who I am, and why you should distrust me a little

I was born in North Africa. I speak French and Arabic as well as English. I run my own company from wherever I happen to live, which at the moment is Cebu. It is American on paper, small, and spread across several time zones. I will stay vague about what it does, because the people I compete with read the internet too. It is enough to say that it lives on words, and that a wrong one costs money.

I read the London Review of Books and Le Monde. I like Houellebecq and Grass, which should tell you that I am not here to cheer anyone up. I lift. I collect watches, the mechanical kind that lose a few seconds a day while my phone keeps perfect time, and I have never once considered the trade. Nobody buys a mechanical watch for its accuracy.

You should distrust me a little because I sell things for a living, and this page is one of the things. Philip Roth said that nothing bad can happen to a writer, because everything is material. The power cuts were material. The invented quotations, the summaries that quietly reversed what the original said, the French that read like a brochure: all of it was material. It became this course.

What the course will be, and what it will not be

It will not be a list of prompts. It will not be a route to ten thousand dollars a month. It will not come with a certificate, and it will not cheer.

It will be short. It will be built for the places I live in and sell into, which the industry ignores: the Philippines, Malaysia, Madagascar and francophone Africa. It will teach a method rather than a tool, because the tools change every quarter and the method does not:

• how to make the machine argue with you instead of agreeing with you;
• how to check what it tells you before you repeat it to a client, a professor or a customs officer;
• how to learn a subject with it rather than letting it learn the subject for you;
• how to write with it so the result sounds like a person who has read a book;
• how to tell which half of your own job sits on the automated side of the Stanford line, and move the rest across;
• how to do all of this on a cheap phone and a connection that drops.

Every lesson will be built around a real piece of work: a report for a client, a grant application, a clause in a contract, a cover letter, a reply to a supplier. No theory without a task attached.

The price is not set. When I set it, it will be a price that makes sense in Antananarivo and Cebu, not in San Francisco.

What you get if you leave your email

• A letter, from time to time, with one worked example: a real task, the machine’s first answer, where it was wrong, and what I did about it. I write when I have something worth your time, not to a calendar.
• The first lessons by email, free, as I write them.
• First access when the course opens, before I spend a cent telling anyone else.
• Now and then, a product I have chosen with care and may earn a commission on. I say so each time. Your address is not sold, rented or shared.

Three objections, answered

I can learn this on YouTube. You can. Most people who try stop after a week, or learn their habits from someone whose actual business is selling the video. What I am offering is the time I have already spent, and a reader who has done the work from your side of the map.

I do not trust AI. Good. Neither does Eric So, and he wrote the book. Distrust is the raw material of this course, not an obstacle to it.

My job is one of the ones being automated. Then you are the reader I had in mind. Stanford found the decline concentrated where AI replaces a task and absent where it helps a person do one. Moving your work from the first column to the second is a skill, and it is the skill this course is about. I cannot promise you a job; nobody honest can. I can show you where the line runs through your own work and how to walk across it.

The screen, the book and what is left

In 2009 Philip Roth predicted that within twenty-five years the reading of novels would be a cultic activity, something for a few thousand people, as Latin poetry is now. The book can’t compete with the screen, he said, and added that twenty-five years was him being optimistic. Seventeen of those years are gone.

Whether he turns out to be right in Tana or Cebu or Penang will depend less on the screens than on what we decide to do in front of them. So’s research says the deciding does not happen by itself; gravity decides, unless somebody designs against it. Villiers’s line cuts the same way: a problem prolonged is a problem made insoluble. That design, for people like us, in places like these, is the only thing I am selling.

Leave your email below. Read the letters. Decide later.

P.S. The course does not exist yet, and I would rather say that here than let you find out on an invoice. I am writing it in the open. The people on this list will see the drafts first, correct my mistakes, and get in before anyone else.