What Actually Changes with AI

Those who lived through the digital transformations of the last three decades from the inside — and I count myself among them — have good reason to greet the current wave with a certain weariness. They have already heard, at regular intervals, that a rupture was about to change everything. Business computing in the 1990s was supposed to reshape the corporation. The internet, in the 2000s, was supposed to abolish distance and redistribute value. Mobile, in the 2010s, was supposed to put immediacy in every pocket and remake our relationship to time. Each time, the rupture kept part of its promises, broke others, and ended up absorbed by a system more resilient than its gravediggers had anticipated.

That weariness is understandable. This time, though, it also risks a misreading.

This chapter rests on a wager: what generative AI does does not belong to the sequence opened by the PC. What remains is to show this, and then draw out the consequences for a region like Valais if it delays taking a position.

Three transitions, one pattern

Let's take these three transitions again and look for what they share.

Business computing in the 1990s equipped functions that already existed — accounting, payroll, sales management, logistics — and made them faster, more precise, more traceable. Bookkeeping work did not disappear; it shifted toward higher-value tasks, and people had to learn the new tools. Geography, for its part, barely moved. The metropolises that housed head offices captured the new computing jobs, and the mid-sized towns that had missed the previous industrial shift rarely found a second chance in digital technology.

The internet, in the 2000s, sprang other locks: access to information, the distance between buyer and seller, the constraints of distribution. It redistributed value, and the print press, travel agencies, and specialty retail paid the price. But to whose benefit? Almost without exception, to players who were themselves urban, financed by concentrated capital, fed by talent pools that already existed. The "end of geography" promised by the web's early prophets never arrived. The opposite happened: San Francisco, New York, London, Paris, Berlin, Beijing, and Singapore captured the bulk of the value created, and the regions that dematerialization was supposed to liberate saw very little of it reach them.

Mobile, in the 2010s, extended the movement by adding immediacy and geolocation. It gave rise to a platform economy — Uber, Airbnb, delivery services, dating apps — that upended entire sets of habits. Here too, value concentrated in a handful of hubs. The smartphone spread across the whole planet; the profits it generates, far less so.

Three ruptures, one logic. They let us do what we were already doing better, cheaper, or at greater scale. The gains in efficiency, reach, and fluidity are undeniable. But the cognitive content of skilled work remained stubbornly human: a lawyer analyzing a contract, a consultant drafting a report, a doctor making a diagnosis, a developer writing a line of code, a fiduciary advising a client. The machine sped things up, sorted, transmitted. It did not produce. Skilled labor remained a scarce factor of production, geographically concentrated, and structurally advantageous for the regions that possessed it.

What generative AI does differently

Generative AI breaks with this pattern, and that is precisely where the stakes become territorial.

For the first time in the history of cognitive tools, a substantial share of the intermediate output of skilled work has become producible by the machine itself. Not final judgment, not accountability, not the decision that commits — on those three fronts, the human remains indispensable, and will remain so for a long time. But everything that sits between the initial brief and the final decision — document analysis, drafting a first version, summarizing the literature, running scenario calculations, technical translation, coding standard functions, framing strategic options — is now produced in seconds, at a quality that rivals an experienced professional's.

Earlier transitions equipped the worker; this one displaces part of what the worker used to do himself. He does not disappear. He directs, validates, adjusts, decides. But the relationship between what he produces and what he has produced changes in nature, and once you add up every skilled worker concerned across a given region, it is the structure of the intellectual labor market itself that shifts.

Two positions nonetheless dominate the public debate, and both seem to me mistaken. The first announces a massive replacement of skilled labor; it requires ignoring everything that engagement, accountability, physical presence, negotiation, and situated judgment continue to protect. The second insists that nothing really changes, that this is just one more tool, that we have heard such announcements before. That is the costlier error: the one made by regions that wait for it to pass. Something is changing. Not everything, but something structurally new — and it touches precisely the geographic dimension of the skilled economy.

The dilution of cognitive capital

For a century, bringing together a critical mass of skilled minds in one place amounted to a nearly unassailable geographic advantage. Talent attracted talent, companies followed the talent, schools were built where demand required them, and the cycle fed itself. What economists call agglomeration effects — knowledge spillovers, specialized labor markets, peer networks, the fertile accidents of chance encounters — gave the great metropolises a multiplier that nothing seemed able to offset.

Generative AI erodes this advantage without eliminating it. Agglomerations keep their pull on the sharpest talent, on capital, on creative milieus; that will not shift anytime soon. What does shift is the threshold of critical mass needed to produce, from anywhere, skilled work of competitive quality. A single consultant in an alpine valley, well equipped and trained in the tools, now commands a capacity for analysis, drafting, and framing that would have required, five years ago, three or four people gathered in a metropolis. A small fiduciary firm handles files that used to be the preserve of large practices. A mountain doctor, mid-consultation, can draw on a body of literature no hospital library ever offered.

This does not mean that quality produced at a distance matches that of the hubs in every respect. On the most exacting assignments, the most situated, the most relational, it will not reach that level. But it comes close enough for the majority of what the majority of clients need. And that majority makes up the economy of a region.

I call this phenomenon the dilution of urban cognitive capital. Dilution, not disappearance: more of it remains in the hubs than anywhere else, but its relative scarcity recedes, because it becomes accessible, in partial but real form, to regions that never held it in their own right. A nearly exclusive territorial asset becomes a partially portable one. The news passes almost unnoticed. It nonetheless changes the game for anyone paying attention.

The other side of the coin: the competitive shift

This dilution does not play out only between urban hubs and non-metropolitan regions. It also plays out, perhaps more deeply, between high-cost skilled economies — Switzerland, southern Germany, certain Nordic regions — and the competitors that had been contesting the market for mid-value cognitive services for twenty years. This effect gets discussed less. It may matter more for the Swiss economy over the next decade.

For two decades, the skilled-services economy followed a simple logic. High-value functions — project architecture, client relationships, decisions that commit — stayed in high-cost countries. Mid-value functions — writing code, drafting documents, processing data, technical translation, routine modeling — migrated toward cheaper skilled-labor pools: India, Poland, Southeast Asia, North Africa. Alongside this offshoring came a second pressure: standardized software solutions, mostly American, absorbing a growing share of generic needs at costs local players could not match.

On both fronts, the balance of power left no room for ambiguity. An offshore developer earning ten or twenty thousand euros a year mechanically beat a Lausanne-based developer paid over a hundred thousand. An American platform charging twenty euros a month per user crushed the bespoke work of a Valais consultant billing twelve hundred euros a day. Switzerland gave up what it had to give up, and its fabric of skilled services retreated to what neither offshoring nor SaaS could take from it: proximity, trust, situated judgment, legal sovereignty.

That balance of power is now shifting. Used well by a qualified local team, generative AI produces in a few hours what an offshore team produces in several days. It adapts to a client's precise context what a standardized platform can only offer generically. It clears, without effort, barriers — time zones, administrative languages, professional cultures — on which distant subcontracting has always stumbled. The cost differential does not vanish; an offshore developer will remain structurally cheaper than a Swiss one. But it stops being the argument that settles everything.

On projects I have followed within my own group, at constant revenue, the productivity ratio per senior professional has shifted by a factor of roughly four to one on development tasks, and about five to one on design tasks — provided a senior remains present to architect and orchestrate the work; I will return to this point shortly. The drawbacks of distant subcontracting, meanwhile, have not moved: delays, friction, uneven quality, losses on complex specifications. The economic equation rebalances accordingly. On certain segments, it simply inverts. These orders of magnitude are the ones my own direct experience allows me to cite; other players observe different figures, sometimes more modest, sometimes higher, depending on their trade. The direction of the movement, however, does not vary.

Symmetrically, the standardized solutions that captured generic needs lose part of their edge. An augmented local team produces custom work tailored to the client, under economic conditions it had not enjoyed since the industrialization of services. SaaS remains relevant for highly standardized needs. As soon as regional, sectoral, linguistic, or regulatory specificity enters the picture, the advantage of bespoke work reappears.

Let's call this effect the competitive shift, for lack of a better term. It changes the game for an entire category of regions that the previous twenty years had marginalized in the field of skilled services. It reopens, for Switzerland and especially for its cantons with strong human capital, markets once thought lost. On condition of seizing it.

The bottleneck: the intelligence that orchestrates

This condition deserves to be stated plainly, since enthusiastic talk about AI tends to skip over it. The competitive shift does not operate on its own. It requires, above the tools, an experienced human intelligence capable of structuring the problem, orchestrating the production, and guaranteeing final quality. It is this intelligence — not AI itself — that becomes the strategic bottleneck of the new cognitive economy.

Generative AI, as currently used, is powerful and imprecise. It produces a great deal, fast, with quality that varies widely depending on what is asked of it and how. Without human architecture upstream, it generates content that sounds right but can miss the actual need. Without orchestration along the way, it delivers fragments that do not fit together. Without validation downstream, errors slip through that no one else will catch. Three moments, three crafts of judgment.

The regions that will capture the competitive shift are therefore those that gather a critical mass of senior professionals able to fill this role. Graduates fresh out of school who already know the tools remain indispensable; that is not enough on its own. The scarce asset is accumulated experience: the kind that spots a poorly framed brief, anticipates a production error, knows how to translate a client's requirement. And it is precisely this asset that regions like Valais can attract and retain, because what they offer — quality of environment, taxation, rootedness, institutional sovereignty — weighs more heavily for a mid-career professional than for a young graduate just starting out.

This distinction matters, because it overturns the usual diagnosis. AI multiplies the most experienced skills rather than leveling out skills in general. Any territorial strategy that bets solely on youth is probably wrong on that count. For a canton seeking to capture the new cognitive economy, the real wager lies elsewhere: creating the conditions for experienced skilled professionals to settle there, or to choose to stay. They are the ones who unlock the leverage effect.

Three scenarios for non-metropolitan regions

This dilution, and the shift it triggers, do not mechanically favor every non-metropolitan region. They open a fork in the road, one that Europe's economic geography already lets us glimpse in three forms.

First scenario: amplification. Far from being weakened, urban hubs capture most of the gains. Their companies adopt AI faster, their talent learns to use it better, their capital finances the innovations that multiply its uses. Non-metropolitan regions watch a new wave of concentration unfold, more brutal than the previous ones because it runs on near-zero marginal cost. The gap between Switzerland's major centers and the rest of the country widens quietly. At this stage, this is the most likely scenario if nothing explicit is undertaken.

Second scenario, the opposite: diffusion. Regions that equipped themselves in time — robust connectivity, legible tax policy, preserved quality of life, training for the workforce already in place, even a modest peer network, and above all a critical mass of seniors capable of architecting work — capture part of the movement of skilled relocation and of the competitive shift. Experienced executives, entrepreneurs, and skilled independents set up shop there because they can now do so without a significant loss of productivity, and because the market itself is reorienting toward what they know how to produce. This scenario does not erase the metropolitan advantage; it redistributes a fraction of it, sometimes enough to change the trajectory of a canton or a valley. Nothing spontaneous about it. It requires decisions, trade-offs, coordinated investment.

Third scenario, the hardest to spot because it plays out within the category itself: fracture. Not all non-metropolitan regions start from the same line. Some have already decided, or are beginning to decide, and will end up on the right side of the dilution. Others wait, watch, put things off, and will end up on the wrong side — not defeated by the urban hubs but overtaken by comparable neighbors. It is this third scenario that most directly threatens Valais. Not head-on competition with Zurich or Geneva: lateral competition with the other alpine regions chasing the same talent — Tyrol, Haute-Savoie, Vorarlberg, Trentino — regions that will not wait for us to make up our minds. I will return to this in the chapter devoted to demographics⁴.

What remains structurally urban

The analysis would remain incomplete, and hardly credible to a serious reader, if it ignored what AI does not change, or does not change to the same degree.

Agglomerations retain four advantages that AI dilutes only at the margins. The most precious, and the hardest to reproduce elsewhere, comes from the creative ferment that density of peers generates: hallway conversations, chance encounters in an incubator cafeteria or at a private dinner, the weak signals you only pick up by being where they circulate. No remote communication tool reproduces that quality, and generative AI even less so. That this is precisely what holds up best says a great deal about the nature of urban cognitive capital: it depends, more than we think, on the chance of encounters.

Three other advantages hold just as firmly, without requiring the same elaboration. Direct access to capital, to investors, to financial decision-making committees remains materially urban, since head offices and funds are located there. The signaling effect declines, but slowly: being in Zurich, Geneva, Paris, or London still means something to an international client who does not have time to evaluate every provider on merit alone. The educational and research ecosystem, finally — universities, engineering schools, university hospitals — remains geographically anchored and continues to draw a share of the youngest talent.

Acknowledging these advantages weakens this book's argument not one bit. Non-metropolitan regions positioning themselves on AI are not trying to become miniature hubs; they never will be. They are trying to capture, within the ongoing dilution, what can be captured: profiles who no longer need the daily ferment, who have already built their network elsewhere, who prefer quality of life to social visibility, or who simply produce exportable value from anywhere. That is not everyone, far from it. But at the scale of a canton of roughly three hundred seventy thousand inhabitants, it is not nothing either, and it can make the difference between a fate as a showcase and a fate as an actor.

Within this configuration, Valais occupies a singular position. It holds rare assets — institutional, energy-related, geographic, cultural — that make it a serious candidate for diffusion. It carries demographic, climatic, and economic vulnerabilities that expose it to fracture. Amplification, for its part, is already happening, and no intervention from Valais will change that. What remains to be seen is which of the other two options will prevail, and that depends on decisions to be made over the current decade. The next chapter examines these assets and vulnerabilities concretely, tested against the transition just mapped out here.