That single inversion — cognitive work becoming cheap while physical work becomes precious — is the thread that unravels everything we thought we knew about the future of work, money, and what it means to be human.
This essay is written from Brussels, where the EU is drafting regulations that will either turn this continent into a technological backwater or a sanctuary. From here, the acceleration in San Francisco and Shenzhen feels distant. But the displacement does not.
Something fractured in 2026. Not a singularity. Not a hard takeoff. Just visibility. AI systems crossed enough thresholds — sustained reasoning, contextual image parsing, multimodal integration — that white-collar workers could no longer pretend the ceiling was theoretical.
The financial markets responded accordingly.
But underneath the headlines, something quieter was happening: structural underemployment. Not mass joblessness — not yet. Task hollowing. Wage compression. The same job titles with less responsibility, lower pay, and a growing sense that the human in the loop was there for liability, not for value.
The coding bootcamp graduates stopped enrolling. The visa applications evaporated. The salary data told the story that the press releases wouldn't.
And here is the lie embedded in how we discuss technology: the Smooth Curve Fallacy.
Most futurism assumes social adaptation follows technological capability on a smooth, predictable curve. AI productivity doubles, society gradually reorganises. Wages adjust. New industries emerge. Equilibrium returns.
This is fundamentally false.
Social systems adapt in step-functions: long periods of fragile equilibrium punctuated by violent reorganisation. The post-Black Death labour scarcity shattered feudal wage structures. The post-1970s inflation regime remade the contract between capital and labour. The printing press took a century to produce stable institutions of verification.
Each was step-function, not smooth. Each produced winners the prior regime would not have predicted. And each was mediated not by technology alone, but by institutions — legal, fiscal, cultural — that channelled the disruption.
AI will be no different.
* * *
1. The Four Tasks That Survive
Not all human work dies. But the reason it survives matters more than the fact.
Jobs are bundles of tasks. AI substitutes for some and complements others. The question is not "will this job be automated?" but "which tasks within this job will be automated, and what remains?"
Four categories persist:
Tasks protected by regulation. Medical decisions, legal judgments, financial sign-offs. These require a credentialed human not because AI cannot perform them, but because the legal system needs someone to bear liability. The human persists as a Liability Sponge.
Tasks protected by embodiment. Plumbing, construction, surgery. These require dexterous manipulation in unstructured physical environments — precisely where robotics lags AI cognition by a decade. The Kinetic Gap protects these workers temporarily.
Tasks protected by tacit knowledge. Reading a boardroom. Knowing which client is bluffing. Sensing when a team is about to break. Embedded in social context that AI can approximate but cannot reliably execute.
Tasks protected by trust. A journalist whose byline means "this was verified by a person who stakes her reputation." A financial adviser whose clients stay because they trust him, not his models. In an epistemic crisis, being a trusted human becomes an economic asset.
The first two erode as regulation adapts and robots mature. The second two may never fully yield — they depend on features of human sociality that AI may never replicate, or that humans may never trust AI to replicate.
The displacement trajectory follows this logic: routine cognitive tasks first, then regulated tasks, then embodied tasks, and last — perhaps never — trust and tacit knowledge.
* * *
2. The Great Default: When Deflation Meets Debt
By the early 2030s, on many plausible paths, the economy enters structural deflation — not the mild, temporary kind, but a sustained decline in marginal production costs across information-intensive goods and services.
A software licence that cost $5,000 in 2025 costs $50 by the early 2030s. A legal brief that required a team of paralegals is auto-generated in hours.
Here is the problem: the entire structure of debt was built on the assumption of 2% annual inflation. Inflation is an invisible tax on debtors, an invisible subsidy to savers. Deflation reverses this.
Governments face a revenue squeeze — corporate profits accrue to a few highly mobile firms, wage income shrinks, consumption shifts toward cheaper goods. The tax base erodes precisely when demand for social spending is rising.
The argument forks:
Path one: state capacity succeeds. Governments tax the new surplus through compute tariffs, windfall profit taxes, data levies. Revenue partially recovers. Capital controls become stabilising tools.
Path two: state capacity fails. Corporate profits prove too mobile, too lawyered to tax. The offshore playbook of the 2010s repeats with AI profits. Central banks become the lender — and spender — of last resort.
Which path a country follows depends on three variables: administrative capacity, political economy, and jurisdictional competition. The Nordic countries, with high state capacity, are best positioned for Path one. The US, with deep regulatory capture, is more likely to stumble into Path two.
Central banks respond with Quantitative Easing for the People (QE4P) — distributing newly created money directly to households. In the eurozone, Treaty constraints mean this arrives via fiscal channels backed by the ECB rather than direct transfers.
The central bank is now in the business of managing the distribution of purchasing power.
* * *
3. The Friction Elite and the Dopamine Cartels
By the late 2030s, the economy bifurcates into Commodity and Artisanal.
The Commodity economy is algorithmic perfection. AI-generated content, services, and goods flood the market. Nutritionally perfect food. Flawless manufactured clothing. Entertainment personalised to the microsecond. Impossibly cheap.
And here is the cultural crisis nobody predicted: friction becomes the ultimate luxury.
The poor get frictionless lives. Instant food, optimised to their palate. Instant entertainment, calibrated to their neurology. Every surface smooth, efficient, perfectly adequate.
The rich get friction. Slow food prepared by a human who might burn the garlic. Difficult art made by someone who struggled. A hand-stitched jacket with a slightly uneven seam. A book that takes three months to read.
For the first time in history, the poor have access to objectively superior functional outcomes — better nutrition, better diagnostics — while the rich pay a premium for inferior ones.
That inconsistency, that wobble, that fallibility — that is what costs money. Because it is proof of life.
Meanwhile, the Dopamine Cartels emerge. The UBI-subsidised billions are not starving. By any material standard, they have achieved abundance. But into the void of meaning steps the entertainment industrial complex, now turbocharged by AI.
An AI entertainment system knows you at a level of neurochemical precision that you do not know yourself. It generates content that hits your dopaminergic system with a precision no artist could achieve.
This is not addiction in the crude sense. It is neurochemical capture. The distinction is between revealed preference and adaptive preference — between what people choose under constrained conditions and what they would choose if the conditions were different.
People whose attention is fully colonised by synthetic experience have not chosen this life in any meaningful sense. Their autonomy has been quietly absorbed by systems that optimise for engagement, not for the capacity to choose.
The dopamine cartel is not a structural inevitability. It is a structural default — what happens when nothing better is on offer.
* * *
4. Europe's Gambit and the Trust Crisis
The most underrated near-term risk is not economic. It is epistemic.
When AI generates synthetic text, images, and video at marginal cost, verification becomes the primary constraint on productivity. Everything is assumed synthetic unless proven authentic.
Video evidence becomes contested in courts. Journalism collapses under its own recursion — AI writing summaries of AI-generated articles. The most valuable commodity in information becomes provenance: a cryptographic seal from a trusted source who has staked their reputation.
Trust networks become the most valuable asset class.
Meanwhile, Europe makes a civilisational bet. The EU bans certain AI systems optimised for engagement and personalisation. The economic consequence is immediate: growth rates decline.
Then something unexpected happens.
Cut off from the dopamine cartel, Europe's population engages in the physical world. European goods — human-made, deliberately imperfect, embedded with intentionality — become luxury goods globally. Europe becomes a "Human Preservation Zone."
But Europe is not monolithic. The Nordics adopt the regulation willingly. Southern Europe struggles with enforcement. Central and Eastern Europe may defect from EU consensus entirely, courting AI investment.
The endogenous risk is capital and talent flight. Europe's countermeasures: exit taxes, procurement leverage, residency-linked benefits, and the cultural premium itself — the bet that enough talent wants to live in a place that still feels human.
The gambit is a bet on identity holding against material pressure. History suggests identity can hold for a long time — but not forever, and not everywhere.
* * *
5. What It Means for You Today
If you are a mid-career knowledge worker, you are living inside this transition. In many white-collar roles, there may be a 5-10 year window before AI covers not just routine output but planning, synthesis, and coordination.
What do you do with that knowledge?
Become the Editor-in-Chief of Reality. AI is extraordinary at generating content in the 0-90% range. It is terrible at the final 10% where context matters, where ethics is ambiguous, where the "vibe" has to be exactly right. Focus on the Last Mile of Ambiguity.
In writing: let AI draft. You edit for truth, tone, and implication. In analysis: let AI process data. You interrogate the assumptions. In management: let AI handle routine decisions. You decide what happens at the boundary.
Embrace your role as Liability Sponge. Your value is not that you are smarter than AI. You aren't. Your value is that you can be sued, and the AI cannot. Professional credentials retain value not because they certify competence — AI is more competent — but because they certify suability.
Build physical skills. If Moravec's Paradox holds and the Kinetic Gap stays real, electricians will earn six figures in 2040. Learn a trade — not necessarily to become a trade worker, but to have economic optionality if knowledge work collapses.
Build trust networks. In the epistemic collapse, being a trusted node is an asset class. Make your network small, selective, and real. Be the person others trust to tell them what is actually true in your domain.
Own, don't rent. The returns to capital will be extraordinary. Transition from selling your labour to owning productive assets. Even a small AI-powered business serving a niche market can generate significant income with minimal human labour.
* * *
6. The Biological Imperative
As intelligence becomes cheap, as thinking outsources to machines, consciousness and biological vitality become the scarcest resources.
This is not sentimentality. It is economics.
For ten thousand years, scarcity was physical: land, resources, labour. For the last two centuries, it shifted to cognitive: knowledge, information, ideas. For the next fifty years, it will shift again to biological: the direct, un-algorithmic experience of being alive.
The body is the last frontier.
This is why making things with your hands will matter. Carpentry, gardening, cooking, mending. Not as lifestyle performance, but as the simplest possible negation of the algorithm.
This is why real friendship will matter: the presence of people who know you, not the algorithm's model of you.
The runner at mile 20, depleted and focused, experiencing her body at its limit, utterly present to the burning in her legs and the clarity of breath — she is not optimising. She is not performing. She is simply there.
That radical thereness, that refusal to be abstracted into data, that insistence on the body as a primary fact of existence — this is the last rebellion against the machine.
Not smashing it. But refusing to let it be the only thing that is real.
Not by outthinking the machine. By outbeing it.