Some think they are installing a super-spreadsheet, just one more optimisation tool. But they are probably letting a slightly pyromaniac interior designer into their organisations: one capable of rearranging the rooms, but also of tearing up the building's blueprints and turning the floors upside down.
Artificial intelligence is often presented as the next stage of digitalisation. Supposedly it will let us write faster, analyse more data, produce code, translate documents or answer customers automatically.
That reading isn't wrong. It's simply incomplete.
AI doesn't only transform the tools we use. It will change company processes, the organisation of work, the skills we attach value to, countries' positions in international economic chains and, eventually, the mechanisms through which we fund collective solidarity.
So we are not facing a technological problem with a matching textbook solution. We are facing an equation with dozens of unknowns: technological, entrepreneurial, social, educational, economic, fiscal and political.
And even the non-mathematicians among us know this: equations with several unknowns are often the hardest to solve.
First unknown: what happens to the company when the machine takes part in cognitive work?
AI adoption is often described as a fairly simple operation: pick a tool, train the staff and measure the productivity gains.
In reality, no company can seriously exploit these technologies without questioning the quality of its data, its IT infrastructure, its cybersecurity, its decision-making processes and the way responsibilities are allocated.
Who checks an answer produced by a model? Who is accountable when it's wrong? Which data can it be trusted with? Which decisions must remain human? And what becomes of a profession when some of its tasks are automated, but not the others?
Bolting AI onto a badly designed process doesn't necessarily make it intelligent. It may simply let you produce errors faster — which is indeed a form of productivity, just rarely the one you were after.
The most significant transformations may therefore come less from automating an isolated task than from redesigning entire processes. In customer service, for example, AI can qualify a request, look up the relevant information, draft a reply and hand complex cases over to a human adviser. That adviser doesn't necessarily disappear, but their role changes: they become more of a supervisor, mediator or expert. And to take on this new role, they too have to develop their skills and their way of working.
A company that adopts AI in depth will no longer be the same company. Its processes, hierarchies, professions and team composition have to evolve together.
Second unknown: what happens to the knowledge worker?
Joseph Schumpeter described innovation as a process of creative destruction: new activities emerge while gradually rendering certain existing structures obsolete.
The mechanism isn't new. Its target, however, is shifting.
Peter Drucker placed the Knowledge Worker at the heart of the modern economy: the person who creates value by gathering, interpreting and passing on information. Developers, accountants, translators, analysts, engineers, lawyers, teachers, advisers and administrative managers all belong, to varying degrees, to this category.
For several decades, digital tools mainly assisted these professionals. The spreadsheet helped the accountant, the search engine helped the researcher, design software helped the engineer.
AI goes one step further: it can now produce part of the cognitive work itself.
It can summarise a file, draft a standard document, prepare an analysis, generate code, translate a text or handle a routine request. It doesn't necessarily replace an entire profession. But it can automate enough tasks to change how many people are needed, which qualifications are sought and how careers are structured.
The question of junior jobs, which I call “The GenZ-Dilemma”, then becomes central. If the machine does the first analyses and handles the simple cases, how will young professionals gain the experience they need to manage complex situations later on?
A company can boost its productivity today while, driven by the imperative of efficiency, weakening its talent pipeline for tomorrow.
Third unknown: where does Tunisia fit in this new division of labour?
For Tunisia, this transformation carries particular weight.
Part of our integration into the international economy rests on skilled, multilingual and competitive talent: information technology, customer relations, engineering, translation, accounting, outsourced administrative services and a range of intellectual work carried out for foreign clients.
Yet the automation threatening these activities doesn't need to happen in Tunisia at all.
If a European company automates a growing share of its support, development, translation or administrative operations, it may have less need to outsource them. The risk therefore comes as much from decisions taken by our partners as from those taken in our own companies.
But another trajectory remains possible. Tunisian teams equipped with these technologies could produce more, reach new markets and offer more specialised services. AI could weaken standardised activities while strengthening those that demand expertise, trust, domain knowledge or cultural proximity.
To stay with Schumpeter's logic: these scenarios will inevitably coexist, without necessarily benefiting the same sectors, the same regions or the same people.
So the real question isn't whether AI will be a threat or an opportunity. It is what position Tunisia will occupy in the new value chain: producer of solutions, exporter of augmented services, dependent user, or mere consumer of technologies designed and operated elsewhere.
Creative destruction offers no guarantee that what gets created will appear in the same place, for the same people, or at the same pace as what gets destroyed.
Fourth unknown: how do you fund a society in which value creation depends less on human labour?
The transformation of companies and jobs inevitably ends up on the doorstep of public governance.
Our tax and social systems were built around an economy in which human labour plays a central role. Wages, social contributions, income tax, household consumption and corporate profits directly or indirectly fund public services and social protection.
But what happens when a company produces as much — or more — with fewer employees, thanks to models operated from data centres located mostly abroad?
Part of the value then shifts away from local labour towards technological capital, digital infrastructure and intellectual property. And, until proven otherwise, a GPU pays into neither the pension system nor health insurance.
The paradox is obvious: just as technological change could increase the need for training, retraining and social protection, it could also erode a significant share of the resources that pay for them.
This doesn't mean we should immediately tax robots, models or computing power. Poorly designed taxation could penalise local companies without ever reaching the big international platforms. But changing nothing would amount to assuming that rules designed for the industrial economy of the 20th century will remain perfectly suited to an economy in which digital capital produces a growing share of value.
One equation, not four separate problems
It would be tempting to hand AI to the engineers, employment to the economists, training to the universities and taxation to the politicians. Yet this siloed thinking is an inherent part of the problem.
A technological decision transforms a company's processes. That transformation changes jobs and skill requirements. Shifts in employment affect public revenue and social needs. Educational, fiscal and regulatory choices then shape the way companies adopt the technology.
We are dealing with a system of feedback loops: each transformation acts on the others and, in return, absorbs their effects.
So we will need to steer clear of two equally comfortable certainties: that AI will spontaneously solve our problems, and that it will mechanically doom human work.
But just as an equation with several unknowns can have several solutions, several futures remain possible. They will depend less on the abstract performance of AI than on the companies we build, the skills we develop, the gains we choose to share and the institutions we manage to adapt.
This article is the opening of a series that will explore three dimensions of this equation in turn:
1. The entrepreneurial impact: how AI transforms processes, professions, organisations and business models.
2. The societal impact: how it affects the knowledge worker, careers, training, inequality and Tunisia's place in international value chains.
3. The governance puzzle: how to adapt taxation, social protection, regulation and economic sovereignty to value creation that is increasingly produced by digital capital and less and less by human capital.
Possible solutions will come afterwards — not as a single recipe, but most likely as several combinations of complementary measures. Because before you can solve an equation, you first have to identify its unknowns.
And that starts with a thorough diagnosis.
Dr. Sami Bahri
Digital Transformation Expert