So I read the Citrini piece. You probably did too. It was everywhere last week. “The 2028 Global Intelligence Crisis.” AI automates white-collar jobs, those people stop buying things, demand collapses, companies automate harder, repeat until civilisation ends I guess. Markets apparently moved on it.
And look, the core loop is sound. If you fire the people who buy your stuff, eventually nobody’s buying your stuff. They even have a good name for it — “Ghost GDP” — where productivity looks great on paper but the actual flow of money between humans has dried up. The mortgage analysis is genuinely nasty: most household wealth is property, most property is financed against salary assumptions, and those salary assumptions are about to stop being true.
I don’t want to dismiss any of that. It’s a real mechanism.
But I keep getting stuck on two things they never address.
The system was already breaking
First: Citrini treats the current economic system as a stable baseline that AI disrupts. But it isn’t stable. It hasn’t been for a while.
We’ve got a climate crisis driven by incentives that reward pollution and punish prevention. A housing crisis driven by treating shelter as a financial instrument. A food-and-healthcare system so misaligned it’s cheaper to make people sick than to feed them well, and then profitable to treat them once they’re ill. Infrastructure that’s been underinvested for 30 years. A recycling system where an AI audit of a UK facility found 93% of what was being thrown away as “residue” was actually recoverable material, not because recovery isn’t valuable but because nobody built the system to do it properly.
None of this needed AI to go wrong. The incentives have been broken for decades. AI didn’t cause the polycrisis (the technical term for “everything is broken and the broken things are making each other worse”). It’s arriving into one.
Which means the “before” state Citrini is trying to protect doesn’t actually exist. There is no stable status quo to defend. The system needed redesigning whether AI showed up or not.
AI is either the straw that breaks the camel’s back or the repair tool. Possibly both, simultaneously. The question isn’t “should we change the system?” We don’t have a choice. We have to change the system. The question is what we change it to, which opens up a cascade of harder questions. In whose interest are those changes? Do they increase the potential future choices for humanity, or reduce them? And recognising that doing nothing is itself a choice: are we willing to let change come through collapse, or are we going to be proactive about steering it toward something good?
What’s getting cheaper
Okay, so what’s actually changing?
Khan Academy’s AI tutor costs $4 a month. A human tutor is $25-80 an hour. The studies say outcomes are as good or better than a human tutor. Four dollars. In refugee camps in Mali and Chad, AI is now screening for TB. Not “fewer radiologists than we’d like” — zero radiologists. The AI is it.
AI therapy: $75 a year, clinically validated. A human therapist in the UK: north of £1,500 for twelve sessions.
“Yeah but you can’t eat a chatbot.” True. But DeepMind ran one model and discovered 2.2 million new crystal structures. 800 years of materials science. 528 new lithium-ion conductors, which is how batteries get cheap, which is how energy gets cheap, which is how quite a lot of physical stuff follows. AI demand forecasting is cutting food waste in half in industrial kitchens. A hundred thousand Tesla home batteries are being coordinated as one 535-megawatt power plant.
So the cost of essential services, and increasingly physical goods, is collapsing. Economists have a name for the old version of this problem: Baumol’s cost disease. Services like healthcare and education get more expensive over time because you can’t easily automate a human teacher or a human doctor. Except now you sort of can. AI doesn’t just make existing services cheaper. It breaks the mechanism that made them expensive in the first place.
Citrini’s death spiral assumes wages go down and costs stay the same. But if costs are falling too, the floor isn’t the same floor.
Here’s the thing though. “Stuff gets cheaper” is not the same as “problem solved.” Even if healthcare and education cost a tenth of what they do now, people still need income. The mortgage broker who gets laid off in 2027 has a mortgage denominated in today’s money. “Everything’s cheaper” doesn’t help if your income is zero.
The cost collapse buys time and lowers the stakes. It’s necessary but not sufficient. You also need a transition.
What the transition actually looks like
A friend made a point I keep coming back to: “Nobody has ever written a detailed scenario showing the exact steps to a positive future. You’d think someone would have a plan for how this goes well. Nobody does.”
I think that’s wrong, actually. But the pieces haven’t been assembled in one place. Let me try.
Step one: freelance constellations. We’ve already got a massive job crunch — not mass unemployment yet, but accelerating displacement of white-collar work. The obvious pattern: people don’t just sit around waiting for new jobs to be invented. They freelance. The difference now is that AI-mediated coordination means freelancers can co-op and act as a larger entity: a flock of highly qualified consultants that a business can engage like a company, even though it’s a flexible constellation of individuals. Businesses-as-usual still get their headcount problem solved. Displaced workers still earn. Nobody had to wait for policy. This is already starting to happen.
Step two: collective action with teeth. Four-day work weeks. Job sharing. Reduced hours instead of full layoffs. The UK four-day week pilot showed 92% of companies continued after the trial: 65% fewer sick days, 57% less turnover, revenue flat. This isn’t charity, it’s good business, and it buys time.
But here’s the thing: it won’t happen voluntarily at scale. Individual companies face a prisoner’s dilemma. The first to cut hours fears losing ground to competitors who don’t. Governments won’t mandate it because businesses threaten to leave. Businesses won’t act because shareholders demand returns and competitors won’t move first. Voters want cheaper goods. The result is a standoff where the two most powerful groups in society fail to act on things that everybody, privately, agrees are essential.
For over a decade I’ve been arguing for something I call a union of humanity: a collective bargaining structure that exists specifically to break these standoffs. Unions, when well-run, are one of the few institutions that have actually delivered large social gains in peacetime. Not through revolution. Through organised negotiation. The mechanism already exists. What’s missing is the scope and the coordination to apply it at the level these problems demand: coercing a four-day work week across industries, forcing stronger action on climate, creating the political cover for governments and businesses to move simultaneously rather than waiting for the other to go first.
If you can spread the same work across more people with fewer hours each, you reduce the pace of displacement without stopping it. That matters because the cost deflation from step one needs time to feed through.
Step three: the work that needs doing. Here’s the bit that I think most people miss. Nurses, teachers, social workers, carers for the elderly. These are spectacularly important jobs. They are also, in economic terms, spectacularly undervalued. We have an economy that pays hedge fund managers two hundred times what it pays the person keeping your grandmother alive and dignified. Not because the market thinks care is worthless, but because the accounting system has no way to capture what it’s actually worth. The same applies to environmental restoration, infrastructure repair, community building.
There is an enormous amount of work that desperately needs doing and that we’ve never figured out how to pay for. Not because it isn’t valuable. Because the system that prices things can’t see it.
As work weeks shrink from five days to four to three, those freed days become available for this kind of work. As AI makes the expensive stuff cheap, the fiscal space opens up to fund the stuff that was always too expensive to organise. The mismatch between “there’s nothing for humans to do” (wrong) and “there’s no way to pay humans to do the things that need doing” (a solvable accounting problem) is where most of the opportunity lives.
And before the “but people won’t work if you just give them money” crowd shows up: the Stockton SEED experiment gave 125 people $500 a month for two years. Full-time employment went up, from 28% to 40%, versus 5% in the control group. It inspired over 100 similar pilots across the US. Turns out when people have a floor to stand on, they don’t lie down. They get to work.
This isn’t a utopian fantasy. It’s three practical steps, each of which is already happening in some form, that together add up to a transition: freelance coordination absorbs the initial shock, collective action spreads the work and breaks the standoff, and redirected time goes to work that was always needed but never funded.
The accounting has to change
“Great, but who pays for all this?”
Fair. This is where the Ex’tax Project comes in, and the reason I keep bringing it up is that it’s the most concrete, serious answer I’ve found.
Cambridge Econometrics, working with Deloitte, EY, KPMG, and PwC — all four Big Four — modelled a specific fiscal shift across all 27 EU member states. Right now, about 51% of EU tax revenue comes from labour. About 6% comes from pollution and resource use. So we’ve accidentally built a system where it’s expensive to employ people and cheap to trash the planet. (If you did this on purpose you’d be a Bond villain.)
Their proposal: shift it. Tax pollution and resource use more, tax labour less. Budget-neutral: no new money needed, just move where the tax falls.
The results: GDP 1.6% higher. Employment 3.0% higher, meaning six million more jobs across the EU. CO2 emissions down 7.1%. EU fossil fuel imports drop by €56 billion. All within five years.
Six million jobs across a continent isn’t transformative on its own. But this is one policy shift, budget-neutral, modelled by Cambridge Econometrics, backed by all four Big Four. And basically nobody I talk to has heard of it.
This isn’t the final answer. But the core idea, taxing bad stuff rather than good people, proves that when people throw their hands up and say there’s no credible path forward, they’re wrong. There’s a plan on the table with numbers that add up. The barrier isn’t knowledge. It’s coordination.
The coordination problem
Getting governments, companies, unions, and regulators to agree on something this big has always been a nightmare. Tax is connected to welfare. Welfare is connected to property law. Property is connected to trade. Pull one thread, whole sweater.
But — and I keep coming back to this — what if coordination just got cheap?
Here’s what’s already happening, quietly. DeepMind ran an experiment where they asked an AI to design a wealth distribution mechanism. Then they put it to a democratic vote alongside standard left-wing and right-wing proposals. The AI’s version won. Not because it was radical, but because it found a compromise that satisfied more people than any human-designed option had. It could see trade-offs across the whole system that individual negotiators couldn’t hold in their heads at once.
CO2 AI gave a single company precise emissions data across 25,000 products in two months. Work that would have taken years with consultants, done in weeks. Stuff that was invisible is now measurable, and when you can measure something, you can negotiate about it. Quadratic funding uses a mathematical formula to amplify small donations: if a thousand people each give $1 to a project, it gets more matched funding than if one person gives $1,000. The idea is that breadth of support matters more than depth of pockets. It has already moved $67 million to over 5,000 public goods projects. Boardy, an AI that facilitates peer-to-peer introductions, is quietly doing for professional networks what dating apps did for relationships: using machine intelligence to find non-obvious matches between people.
Small examples. But five years ago none of this existed. The cost of getting large groups of people to agree on complicated things is collapsing. So why hasn’t anyone assembled the detailed positive scenario? Not because it’s impossible. Because it requires coordinating an insane number of moving parts, more than any previous generation could handle.
We might be the first that can.
Change the scoreboard
You can’t run a new operating system if everyone’s staring at the old scoreboard. If GDP is your only metric, cheaper services look like shrinking output. A world where AI tutors replace expensive ones, where community care replaces institutional care, where local energy replaces imported fuel. That’s GDP contraction by definition, even if actual lives are getting dramatically better.
This matters because the scoreboard shapes the policy. If governments measure success by GDP growth, they’ll optimise for GDP growth, which means optimising for expensive solutions over effective ones. The metric becomes the prison.
Over 50 cities are now running Doughnut Economics dashboards, measuring whether people have enough without overshooting what the planet can handle. Amsterdam has measurably cut CO2 with one. New Zealand’s treasury is legally required to report against wellbeing indicators. Scotland’s tracking 81 different measures of how the country’s actually doing.
If your scoreboard includes health, education, environment, connection, and access to opportunity, then a world where AI makes essential services nearly free doesn’t look like collapse. It looks like abundance.
And abundance creates space. Space for experimentation, for regulatory sandboxes where new economic models can be tried without betting the whole system. Space for people to start things, to try things, to fail cheaply and try again. The opposite of scarcity thinking. The economy we have is a scarcity machine that creates artificial bottlenecks. The economy we could have is an abundance machine that creates room to move.
Citrini is right that the current system can’t absorb what’s coming. But the current system was already failing. The polycrisis didn’t start with AI and it won’t end by protecting the status quo.
We designed this economy. We chose how to measure value, where to put the tax, what to count. None of it is a law of physics. We made it up, it was useful, and it’s becoming less useful. The question was never “will AI break the economy?” It was always “the economy is already breaking — can we coordinate the update in time?”
There are modelled proposals. There are practical transition steps. There are new coordination tools. There’s even a new scoreboard.
This is the short version. Each of these threads deserves a longer treatment, and I intend to write them. But the outline is here, and the pieces exist.
It shouldn’t be easier to imagine the end of civilisation than the end of an accounting system.