I install AI for clients. A consultancy buried under the same questions by email. A shop that repeats dozens of times a day that it delivers to Adidogomé. Every time, the same scene: nobody loses their job, but someone discovers that part of their day has just changed hands.
That is what we need to talk about. Not the end of work.
Two facts, laid side by side
More than 1,000 layoffs were recorded at African startups in the first half of 2026, against a backdrop of cost cutting. Over the same period, these startups have been turning more and more to artificial intelligence for customer service, fraud detection and creditworthiness assessment.
I leave the two facts side by side without drawing an arrow between them. Cost cutting has a thousand causes, and I am not pointing at anyone. But that list of three uses is worth pausing on: customer service, fraud, creditworthiness. Three places where the work consists of handling a great many cases that look alike.
And the tool is already in people's hands. In the first quarter of 2026, 10.1% of Togolese aged 15 to 64 say they have used a generative AI tool. One in ten. This is no longer a conference topic.
It is tasks that move, not jobs
When I open up a team's real work, I never find one solid block. I find a stack of motions.
The first-level reply. Opening hours, prices, the address, the status of an order. At one client, the bulk of the message volume came down to a handful of questions. That is the first thing to go.
Sorting. Classifying, labelling, routing to the right person. A quick motion, repeated a thousand times, that wears you out without producing anything visible.
Routine checking. A file complete or not, a document present or missing, an amount consistent or not.
Then there is the rest, inside the same job. Understanding why this customer is angry when their order did arrive. Deciding whether to make an exception for a regular. Turning down a file you can feel is shaky without being able to write it up as a rule. Telling someone they got it wrong without humiliating them.
That part, I have never seen move. It is judgement. And judgement needs a context the tool does not have.
What I would do on Monday morning
Write your tasks down for a week. No memory, no estimates. Write, as the day goes, what you do and how long it takes. One week is enough to make the repetitions surface.
Draw two columns. On the left, what follows a stable rule: same input, same output, every time. On the right, what requires weighing up, understanding, refusing. The left-hand column is the one that moves. That is not a threat, it is information.
Automate a single task from the left this week. One, not ten. Questions that keep coming back are handled with a page of written answers and a model. The point is not the technology, it is getting hours back.
Put those hours back into the right-hand column. Otherwise you have just worked faster to do the same thing. The time you save has to go into meeting a client face to face, understanding their business, fixing a process that grinds.
Checking becomes the real work
A tool that answers fast also produces wrong answers, with the same confidence. I have seen an automated reply announce a lead time that did not exist. Nobody had re-read it. The customer, though, had read it.
So the role that appears is not the one that writes. It is the one that re-reads: spotting the plausible error, knowing what the tool usually gets wrong, cutting it before it goes out. That is not learned from a tutorial. It is learned by keeping a log of what went wrong, and reading it back.
Whoever steers the tool and controls its output holds a solid place. Whoever performs the task the tool takes over holds a fragile one. Between the two, there are a few weeks of serious work. Not years of study.
What is improving, honestly
Access has widened. A seamstress who takes her orders on WhatsApp can write a clean quote, translate an exchange with a customer, prepare an ad. Before, she paid someone for that. The ceiling is higher for whoever grabs hold of it.
What is missing is the support. Many open the tool, ask three questions, are disappointed, close it again. The useful motion is not in the tool. It is in the way you cut up your own work before you open it.
It is not the tool that decides my place, it is what I do with my right-hand column.