Beyond the Flywheel - AI as Infrastructure
For twenty years, we've heard the same stories about growth loops turning faster, users bringing more users, data feeding better services, and momentum compounding like physics. The flywheel explained why Amazon and Facebook felt unstoppable and why Uber could burn billions and still look inevitable. We believed in it because it was happening right in front of us, and we watched it grow without ever seeming to stop.
Through bitter experienced we've learned that every flywheel ends the same way. Growth slows and cheap money dries up, so platforms start squeezing. They raise fees and pack in more ads, putting pay-to-play ahead of merit while throttling APIs and organic reach. Service degrades as support shrinks, and new features tilt toward what monetises rather than what actually helps users. There's a name for this: enshittification, the steady shift of value away from users and toward quarterly targets.
Reddit hiked API prices so high that Apollo shut down and entire communities went dark in protest. Twitter killed free API access, breaking the third-party apps people actually liked using. Amazon set Prime Video to show ads by default unless you pay extra. Instagram tested opening straight into Reels, burying the feed people came for. Google's first page is now so packed with ads, AI summaries, and sponsored links that people type "reddit" at the end of queries just to find actual human answers. Each move trades user experience for shareholder numbers.
And it isn't just users getting squeezed. Facebook killed local news ecosystems but never hired reporters. Uber destroyed taxi systems built over decades, replacing worker protections with surge pricing. Google trained its AI on everyone's content, then used it to bypass the creators entirely. They privatised the town square and left us with the bill: anxious teens, precarious workers, broken democracies. The flywheel's genius isn't innovation; it is extraction. Move fast, break public goods, bank the value, leave society to clean up.
This matters now because we're at a choice point. The flywheel's failure shows us what not to build. Don't mistake motion for strength, since it was always temporary, spinning growth as its own defence.
At the precise moment platforms are degrading, artificial intelligence arrives.
The Choice
We can treat AI as another extraction machine for the next earnings call, or we can build computational intelligence as civic infrastructure for the next century: something that holds under strain, expands human capability, and serves everyone.
Infrastructure protects and empowers; platforms just extract. This is nation-building, not algorithms alone, but the workforce to shape them, the trust to implement them, the independent technology stack to run them, and the capital to sustain them. These are foundations, like roads and power and law, on which everything else can flourish.
Who Controls the Architecture⚓︎
When we think about building foundations, the question isn't whether AI will reshape society. It's who decides what can be done and how choices get made. Every automated decision shifts power from humans you can challenge to systems you can't. Search results that once offered ten sources now give you one answer. Loan decisions once made by bankers you could appeal to now come from models that offer no recourse.
Soon it won't just be decisions. Agents will act: trading, hiring, negotiating, building, with authority we grant but can't revoke transaction by transaction. They'll move faster than oversight, coordinate beyond human comprehension, and create facts on the ground before anyone can object.
We're not building tools in any traditional sense. We're building decision infrastructure: the computational substrate through which collective intelligence emerges and coordinated action happens. When agents make millions of micro-decisions per second, when they negotiate and coordinate at scales humans can't track, they become the architecture of choice itself. Not tools we wield, but the medium through which economic and social decisions flow.
From Physical to Computational Sovereignty⚓︎
In industrial economies, sovereignty meant controlling physical infrastructure: ports, rail, energy grids. In post-industrial economies, it meant controlling information flows. Now we're entering something different: an era where intelligence itself is infrastructure. Where the capacity to decide, to act, to coordinate at machine speed becomes the foundation everything else runs on.
Look at societies that actually last. They're not built on compounding loops. They're built on the strength of their fabric. Singapore calls this Total Defence: five layers working together to secure the nation. The Nordics talk about Strong Societies, where trust and cohesion matter more than any fortress. The lesson is clear: sovereignty isn't one moat or one product. It's the strength of people, institutions, and infrastructure working together.
We need to treat AI at this level. Not as a quarterly growth engine, but as civic infrastructure that has to be guarded and renewed. Like any critical infrastructure, it needs redundancy, maintenance, workforce development, and governance structures that outlast political cycles. That means four things:
The Four Pillars⚓︎
Workforce
AI will eliminate jobs. Lots of them. But that's not the whole story.
Every major technology kills some skills while birthing others. Telegraph operators vanished, but network engineers emerged. Typists disappeared, programmers appeared. Today's pattern: coders spend less time writing, more time reviewing what AI generates. The skill migrates upward—from production to supervision.
The danger is drifting from "humans in the loop" to "humans on the loop"—from actively engaged to rubber-stamping what the machine decided. That's when expertise dies.
We need reserve skills. The Navy brought back celestial navigation after years of GPS dependence. Not everyone needs a sextant, but someone better know how when satellites go dark. Same principle: enough humans must understand these systems deeply enough to catch failures, to know when the model has lost the plot.
This means real reskilling programs, not theatre. Safety nets that actually work. And preserving what makes us irreplaceable: taste, judgment, the ability to wrestle with ambiguity, knowing when the machine is confidently wrong.
Trust
Responsible AI without enforcement is just performance. If citizens live under systems they cannot challenge, appeal, or scrutinise, legitimacy collapses.
Trust requires audits with real teeth. It requires rights of redress that actually function, not complaint forms that go nowhere. It means confronting failures in public rather than hiding them behind PR statements.
And the harder part: regulation must be shielded from capture. When the people writing the rules are planning their next job at the companies being regulated, the whole system becomes theatre. We need regulators with genuine independence, real technical capacity, and the authority to act when things go wrong. Anything less is just paperwork.
Technology
A nation that depends entirely on external platforms might imagine it's buying efficiency. In truth, it's mortgaging sovereignty.
To rent your stack is to rent your future. If you cannot walk away from a vendor in thirty days, you're not a customer. You're a colony. Every dependency that cannot be broken isn't infrastructure. It's submission dressed up as convenience.
This doesn't mean building everything from scratch or rejecting all foreign technology. It means ensuring you can switch. That you understand how critical systems work. That you have alternatives when you need them. Sovereignty is about optionality: having the technical capacity and workforce to maintain, modify, and replace systems that matter.
The test is simple: can you leave? If not, you don't own your infrastructure. You're renting it, and rent always goes up.
Economy
It's pleasant to say AI will grow the economy. The harder truth is that AI will run the economy at machine speed.
Agents will approve loans, set prices, negotiate contracts, allocate credit, moving faster than human oversight can match. In this world, decision-power itself becomes money. Control over these systems is economic power.
If those decisions are made by systems controlled elsewhere, you haven't just outsourced technology. You've outsourced command of your economy itself.
Think about what that means. When an algorithm somewhere else decides who gets capital, at what price, and on what terms, that's not a technology question anymore. It's a sovereignty question. When foreign systems decide which businesses can access credit, which sectors get investment, which regions get development, you're no longer running your own economy. You're asking permission to participate in it.
This is why the technology pillar and the economy pillar are inseparable. Economic sovereignty requires technological independence. You cannot have one without the other.
Nobody makes a pencil from scratch anymore—hasn't for decades. The cognitive division of labor is old news. But AI cranks this interdependence to warp speed. When agents execute millions of micro-decisions per second, human oversight has to evolve. We can't review every decision. We have to set boundaries, encode values, maintain accountability for what matters.
Economic sovereignty in the age of AI isn't about controlling every transaction. It's about preserving human judgment over the transactions that define your economy's character. Let the machines handle the routine. Keep humans in charge of the exceptions, the ethics, the edges where judgment matters more than speed.
These four pillars (workforce, trust, technology, economy) aren't separate problems. They're facets of the same challenge: building AI infrastructure that serves the public rather than extracting from it. Infrastructure that defends essential capabilities and expands what people and institutions can do.
The flywheel era taught us what happens when we optimise for growth above all else. Now we have a chance to build something different.
Building for Generations⚓︎
What's possible: Systems that coordinate millions of micro-decisions at machine speed while keeping humans in command of what matters. Teachers who actually know when a student is struggling because the system handles the compliance paperwork they used to drown in. Justice infrastructure that learns what reduces reoffending across decades of cases, while judges and caseworkers keep discretion over individual lives. Local councils that can fix potholes and approve permits faster than committees can meet, but citizens still control the budget. Employment services that match people to real careers, not just tick placement boxes for government contracts. Not platforms extracting value from every interaction, but infrastructure that amplifies human judgment at the speed and scale modern society requires.
The easier path is to rent everything, chase the quarterly number, and let platforms extract value until they degrade. That path is well-worn. We know exactly where it leads.
The harder path means building when it's easier to buy, thinking in decades when quarters are what get measured. It means treating AI as infrastructure we maintain not software we subscribe to.
The Nature of Infrastructure
The thing about infrastructure: when it works, it disappears. Roads don't get credit for the economy they enable. Power grids don't get celebrated for the lives they improve. They just work, year after year, holding up everything else.
That's the choice. Not AI that dazzles for a quarter then degrades, but infrastructure so reliable it becomes invisible, so foundational it expands what's possible for everyone who uses it.
The platforms taught us that extraction is temporary. Sooner or later, you run out of things to squeeze. But infrastructure compounds differently. Every year it stands, it enables more. Every crisis it survives, it proves its worth. The work of building it is hard and the returns are slow, but what gets built becomes the ground everything else stands on. That's not a quarterly story. That's how societies last.
The Stakes of Stewardship
Infrastructure is only as good as the humans who maintain it. If we drift from "humans in the loop" to "humans on the loop"—from active engagement to passive approval—we risk what Sartre feared: becoming the machine's machine.
The hardest part isn't the technology. It's maintaining our agency: staying sharp enough to supervise, wise enough to intervene, and human enough to know which skills we can't afford to lose. As Appiah reminds us, if there's one skill that matters above all, it's the skill of knowing which of them matter.