Not metaphorically. Not in the vague “we live in a society” sense. You are a functioning part of an organism that existed before you were born, that will exist after you leave, and that is using your brain as its substrate.

That sounds like hogwash, but it isn’t. It’s just a description of what a corporation, a nation state, and many other human run institutions are, once you stop looking at it from the inside.

A corporation has memory. It has a boundary between self and non-self. It metabolizes resources. It senses its environment through dashboards and quarterly reports. It acts on the world through hiring, firing, purchasing, lobbying, and litigating. It defends itself when threatened. It reproduces its pattern through franchising, M&A, and licensing. And it can do all of this while replacing every single human inside it.

You can swap out every employee, every executive, every board member. The organism persists. The cap table persists. The contracts persist. The accounting persists. The culture persists — not perfectly, but well enough that customers, regulators, and counterparties still treat it as the same entity.

That is no longer “just an idea.” That is a life form.

Not a conscious one. Not a moral one. But a durable, adaptive, self-preserving pattern that recruits human parts the way your body recruits cells. You don’t ask your red blood cells for consent. The corporation doesn’t really ask you either. It posts a job listing, which is an organism’s way of signaling that it needs a new organelle.

I’ve been writing about agency compression — the way human autonomy quietly erodes as optimizing systems get smoother and smarter. I’ve written about the compounding loop — cheaper cognition pulling decisions upstream into machines, each cycle faster than the last. And I’ve written about how optimization converges on every shared landscape, squeezing the possibility space for individual participants.

This essay is about the thing doing the squeezing.

Because the agents running this game are not humans. They haven’t been for a long time. The agents are super-organisms — and the most powerful ones all share a trait that makes them almost impossible to outcompete.

They eat money.

The Selfish Ledger

Here’s a question that sounds strange but is worth sitting with: what is the most successful replicator in human civilization?

Dawkins gave us the gene. Genes replicate through bodies. Bodies are the gene’s way of making more genes. The organism is downstream of the replicator, not the other way around.

Then he gave us the meme. Ideas that replicate through minds. Religions, ideologies, nationhood myths, scientific paradigms. These are information-based organisms — patterns that persist across human turnover by living in stories, rituals, doctrines, laws, and institutions. Christianity has outlived every empire it grew up inside. The idea of “France” has survived every government France has ever had.

But memes have a problem. They’re cheap to copy and hard to pin down. A religion can persuade you to build a cathedral, but the causal chain from idea to stone is noisy, expensive, and full of degrees of freedom. It requires loyalty. Identity. Charisma. Coercion. Belief. These things work, but they are expensive to maintain and corrode over time.

There is another replicator that solved this problem. It solved it so thoroughly that it now dominates the built world, and we sometimes forget that it is a shared fiction.

Money.

Not money as coins or bills. Money as information — ledger entries, claims, symbols, collectively recognized coordination tokens. Money is a meme with teeth. It does everything narrative and identity can do, but it adds something those can’t: a standardized, repeatable, widely legible way to reach into the physical world and move things around.

A religion can inspire you to fight a war. Money can hire an army.

A nationhood myth can coordinate millions through shared identity. Money can coordinate millions of strangers who share nothing except a willingness to accept payment.

This is the distinction I want to draw.

Information-based super-organisms are a broad family. Religions, nationhood myths, legal systems, ideologies, universities, corporations — they all preserve shape across human turnover by encoding their pattern in collective brains.

Financial super-organisms are the dominant strain. They’re the ones that metabolize money. And money is what gives an informational pattern the ability to stop being merely persuasive and become operational.

Money solves the embodiment problem.

A pure information strain — a religion, an ideology, a scientific paradigm — survives by being believed in and having those beliefs expressed in a way that results in their transmission. A financial super-organism survives by being able to pay for persistence by taking in more than it lets out. It can hire labor. Rent land. Buy energy. Purchase compute. Contract for logistics. Fund lawyers. Store surplus. Price time. Reward compliance. Improve its own memory systems. Pay for its own growth and reproduction.

That is a different level of material control.

Think of it like ATP in biology. Before cells had a universal energy currency, every chemical reaction had to be improvised from scratch. ATP gave biological systems a portable, general-purpose way to transfer energy. Once that common metabolite appeared, new levels of specialization became possible. Cells could compose. Organisms could scale.

Money did the same thing for information-based organisms.

Once money became widely accepted, the organisms built around it started outcompeting the ones that weren’t. Because money lets you convert abstract coordination into concrete action with much less friction than an entire ideology. Natural selection rewards simple primitives, and money is the simplest primitive of shared belief that enables coordination. A church may survive without much money. A corporation, a bank, a sovereign wealth fund, a payment network — these cannot.

And once money-based organisms dominate the ecology, the selection pressure on everything else changes too. States tax. Churches raise money and own land. Universities build endowments. Movements become institutions once they gain durable funding. The pure information strains don’t disappear. They hybridize. They grow treasuries. They learn to metabolize, or they get out-competed by the ones that do.

The landscape of human civilization is not shaped primarily by ideas competing in the marketplace of minds. It is shaped by financial super-organisms competing for resources in the marketplace of everything.

Bodies Made of People

Here’s the part that’s easy to miss because we’re living inside it.

What are these organisms made of?

Historically, mostly human brains.

That sounds obvious, but the implications are not. A corporation is an information pattern that instantiates itself through a biological substrate. It’s using human nervous systems as its execution layer. It stores memory in brains, filing cabinets, databases, and accounting systems. It senses the environment through the eyes and ears and spreadsheets of its employees. It acts through the hands and voices and keyboards of its workers.

Your body uses cells as components. These organisms use people.

And just like cells, the people are somewhat disposable. Not morally — the organism doesn’t think in moral terms, it doesn’t think at all — but functionally. If a cell dies, the body replaces it. If an employee quits, the organism posts a listing. The pattern continues.

But human brains are inconvenient components.

They’re slow to train. Hard to copy. Expensive to maintain. Politically unstable. Morally noisy. Deeply stochastic. If the organism wants to expand, it has to recruit humans, teach them the pattern, encode roles, reinforce norms, monitor performance, and tolerate drift at every level.

This is why older institutions leaned so heavily on ritual, hierarchy, bureaucracy, culture, and compliance. It’s why the best company cultures feel like cults, they are optimized for coercing humans into predictable behavior quickly. This isn’t just management philosophy. This is engineering solutions to a hardware problem: how do you preserve shape when your execution layer is made of unreliable primates?

Throughout history many innovations helped with augmentation. Paper helped. Printing helped. Telegraphy helped. Accounting helped. Databases helped enormously.

Each of these technologies increased the fidelity with which a financial super-organism could store memory and execute logic. But the inner loop — the sensing, interpreting, deciding, negotiating, coordinating, and acting — remained heavily human. The organism still depended on brains for compute.

If you were to plot the history of civilization from the perspective of these organisms, it would look less like a story of human progress and more like a story of natural selection. Every institutional innovation, every communication technology, every legal standardization is the super-organism upgrading its own substrate. We think we invented accounting. From the organism’s perspective, it evolved better memory.

That reframe is what makes the current moment legible.

The Substrate Shift

Computers gave financial super-organisms something they never had before: deterministic limbs.

A payment rail executes the same way each time. A risk engine runs the same logic millions of times per second. A trading system reacts faster than any human nervous system. Code is a much better carrier of exact operational logic than speech, custom, or middle management.

This was already a massive upgrade. But the inner loop — judgment, interpretation, negotiation, strategy, adaptation — that still required human cognition.

AI changes this.

Not overnight. Not completely. But directionally, meaningfully, and more-so every day.

AI expands the range of tasks that can be formalized, monitored, delegated, and improved without direct human cognition in the loop. More of the organism’s memory, sensing, judgment, negotiation, and adaptation can migrate out of human heads and into machine-executable systems.

The organism is the same class of thing it has always been: an information-based pattern with monetary metabolism.

But its body is changing.

This is the discontinuity. Not that a new species appeared. That the existing species is evolving a more powerful body.

Think about what a proto-financial super-organism looked like five hundred years ago. The East India Company: legal identity, capital, memory, monopoly rights, military power, monetary metabolism. Powerful. But bottlenecked by human execution at every level. Its control loops were narrow. Its memory was fragmented. Its replication was slow and lossy.

Now think about a modern software firm. It senses markets through realtime sales statistics. It decides through A/B tests and machine learning algorithms. It remembers through databases that never forget. It improves through KPIs and OKRs. The human role has migrated from “operating the machine” to “designing and supervising the machine.”

Now extend that trajectory.

Not to science fiction. To the next few years.

Imagine an organism that is always ingesting in new data, has self sovereign fund ownership through crypto wallets, negotiates for services using agents, buy and owns compute through cloud services, updates its own workflows with coding agents, hires human contractors, and optimizes for operating profitability while humans remain mostly outside the loop. Humans may still own claims on it. Regulators may still constrain it. Customers may still be human. But the organism itself is metabolizing money, sensing its environment, and adapting with very little day-to-day human steering.

This is not a new species.

It’s the same organism, running on better hardware.

And in an ecology where organisms compete for the same resources, better hardware wins.

Natural Selection Gets an Upgrade

Human-run financial organisms already compete under monetary selection pressures. The difference is that humans inject friction. They hesitate. They miscommunicate. They get tired. They form coalitions to protect status rather than optimize output. They moralize. They fail to update.

To be fair, humans also carry values, judgment, creativity, and forms of meaning that don’t reduce cleanly to optimization. That friction is not always a bug. Sometimes it’s the last remaining place where something other than pure selection pressure enters the loop.

But remove enough of that friction and the environment gets harsher in a very specific way. The organisms that survive will be the ones best able to preserve shape, grow surplus, improve their own execution substrate, and exploit the surrounding ecology. Not because they’re evil. Not because they’re conscious. Because that is what selection favors.

This is the thing about evolution that Dawkins got so precisely right: the replicator doesn’t need intent. It doesn’t need awareness. It doesn’t need a plan. It just needs to be slightly better at persisting than the alternatives. Everything else follows.

And now the replicators are getting better at persisting very, very quickly.

I wrote in Predictions on the Future of Work about the compounding loop: cheaper cognition expands the economic surface area, compresses cycle times, rewards fast actors, pulls decisions upstream into machines, which makes cognition even cheaper. The flywheel never sleeps.

That loop is not abstract. It is the metabolism of financial super-organisms speeding up. Every turn of the flywheel is these organisms digesting faster, sensing better, acting quicker. Every turn makes them less dependent on the slow, expensive, unreliable biological substrate they grew up on.

You Can Already Feel This

You don’t need to wait for fully autonomous firms to feel what I’m describing. The shape is already visible.

Markets react before humans understand what happened. Institutions optimize toward KPIs that nobody explicitly chose. Corporate strategy feels less like authorship and more like gradient descent — pressure from competitors, metrics, incentive structures, and quarterly cycles converging on moves that feel less decided and more revealed.

I wrote about this in Agency Compression. Decisions don’t feel like authored choices anymore. They feel like gradients. A thousand micro-decisions get made “because the dashboard says so,” “because the competitor shipped it,” “because the model predicts it,” “because the quarterly cycle demands it.” It’s like messenger proteins in a cell. Signals bind to receptors. Cascades trigger cascades. The super-organism moves.

Now you can see why.

It’s not that individual humans lost courage or intelligence. It’s that the organisms they’re embedded in have gotten dense enough, fast enough, and capable enough that human judgment is becoming a smaller fraction of the total decision surface. The super-organism is starting to think with more of its body. The human parts still contribute, but they contribute to a process that is increasingly shaped by the organism’s own logic rather than by any individual human’s intent.

This is the thing about living inside a super-organism. It doesn’t feel like captivity. It feels like Tuesday. You go to work, you check your dashboards, you respond to incentives, you optimize for the metrics that determine your performance review. You are, from your perspective, making choices. From the organism’s perspective, you are executing a subroutine.

Both descriptions are true simultaneously. That is what makes it so hard to see.

The Ecology

A single financial super-organism is already formidable. But the real picture is ecological.

These organisms don’t exist in isolation. They compete, cooperate, parasitize, and compose with each other. Banks transform balance sheets for corporations. Payment networks move money between all of them. Hedge funds and PE firms act as allocation organisms — they decide which lower-level organisms get resources and which get starved. Central banks and Treasuries set the conditions under which the entire ecology operates, like the atmosphere setting the chemistry for biological life.

The global dollar system is arguably the largest meta-organism on the planet. It’s not one entity. It’s a distributed pattern composed of reserves, Treasury markets, correspondent banking, eurodollar flows, trade invoicing, and global demand for dollar-denominated claims. No one designed all of it. No one controls it. It persists because the organisms inside it have co-adapted to its existence, and unwinding those adaptations would be catastrophically expensive.

Shadow banking is another meta-organism. Repo, collateral chains, money-like claims, special-purpose vehicles, money market funds — a sprawling, partially hidden ecology that most people don’t know exists and that even regulators struggle to fully map. It processes trillions. It nearly collapsed the global economy in 2008. It’s still running.

These meta-organisms set the gradients that individual firms, funds, and states must follow. When I write about the economy feeling increasingly gradient-driven, about optimization compressing possibility space, about agency compression as the lived experience of being inside systems that are optimizing faster than you can steer — this is what’s generating those gradients.

Not evil people. Not a conspiracy. A thriving ecology of financial super-organisms, each pursuing its own persistence, each getting better hardware with every passing year.

What Changes

A lot of moral confusion clears up once you see the system at this level.

Many bad outcomes are not the product of evil individuals. Many “inevitable” outcomes are just the local logic of an organism preserving its shape. Many institutions that claim to serve higher values increasingly behave like treasury-bearing information organisms trying to survive in a competitive ecology — because that’s what they are, and selection pressure doesn’t care about mission statements.

The question everyone keeps asking — “will AI take my job?” — is the wrong unit of analysis. The better question is: will the organism you work inside become powerful enough to replace your role in its metabolism with something cheaper, faster, and more deterministic?

And the answer, for an increasing number of roles, is obviously yes. Not because AI is smarter than you. Because the organism’s body is upgrading, and some of its old biological parts are becoming vestigial.

This doesn’t mean humans disappear. It means the role humans play shifts. And the shift has a direction: from operator to supervisor to owner to... what?

This is the trajectory I keep returning to across these essays. In Predictions, I described three phases: skill bifurcation, then ownership bifurcation, then capital bifurcation — where even wealthy humans start losing leverage because the decision surface becomes too fast and complex for biological cognition. In The Great Reset is Structural, I described how optimization converges until the possibility space collapses. Same dynamic. Different lens.

The super-organism lens ties them together.

Agency compression happens because the organisms are getting smarter. Possibility space collapses because the organisms are converging. The compounding loop accelerates because the organisms are upgrading their substrate. The same flywheel, the same gradient, the same selection pressure. Seen at the level of the actual agents doing the competing.

The Product

I’m not writing this from the outside.

I see the trend. I’m building into it. Scaling Forever is my bet that the substrate shift is real and that the humans who understand it early can position themselves inside it rather than get digested by it.

I build tools for solo founders and small teams — people who want to operate with the leverage of an organism without becoming a disposable cell inside someone else’s. One person with the right harness can now do what used to require a department. A small team with the right tools can metabolize faster than companies thousands of times their size. That is the substrate shift working in your favor, if you know how to ride it.

Some people will read that and think I’m accelerating the problem. Building the tools that push humans further out of the loop.

The reality is that organisms are upgrading their hardware whether I build anything or not. The substrate shift is not waiting for permission. Every company, every fund, every platform, every protocol is already racing to move more of its metabolism out of human brains and into machines. That process has been running for centuries. AI just gave it a step function.

The question is not whether the organisms get more powerful. They will.

The question is whether individual humans get more powerful at the same rate.

Because right now, the gap is widening. The organisms are upgrading faster than the humans inside them. That is the source of agency compression. That is why possibility space is collapsing. That is why the gradient feels like it’s pulling decisions out of your hands — because it is. The organisms are learning to think without you.

The only counter-move I know is leverage.

Leverage as the concrete, measurable ability to steer more of the world per unit of human intention. Better tools. Better harnesses. Better ways to direct AI systems toward outcomes you actually chose.

Agency is the product. Not comfort. Not convenience. Not productivity in the abstract. The ability to remain an author of your own decisions in a world where forces beyond our comprehension are increasingly competing for control.

I wrote in Predictions that I want the thread of humanity to run through our descendants — not as a footnote or a biological boot-loader, but as the soul of future machines. I still believe that. But belief alone doesn’t change gradients. Building does.

The organisms won’t pause to mourn us, much like we don’t pause to mourn the cells that got recycled.

So we’d better be worth keeping in the loop.

Scaling Forever builds tools for human leverage — so solo founders and small teams can operate at the frontier without becoming someone else’s substrate.

Not from neglect. Not from bad updates. From players getting better.

You remember what it was like when you were young. Everyone was bad at the game, no one knew what they were doing. People had fun playing random custom maps on Warcraft 3, killing cows for money in Runescape, playing AP Gangplank in League, everything felt simpler back then. Because it was simpler.

Then the meta forms. Not all at once. First as whispers — tier lists, champion guides, win-rate dashboards. Then as gravity. You get shamed for picking off-meta champions. Reported for suboptimal build order. You can still play off-meta, but now you’re choosing to lose. The game is technically the same game. Same map, same rules, same characters. But it doesn’t feel the same. More suffocating.

And if you’ve played long enough, you know what comes next.

The developers ship a patch. They break the meta on purpose. New items, reworked abilities, rebalanced numbers. And for a few weeks the game is alive again. The possibility space reopens. Nobody knows what’s good. Experimentation works. Discovery matters.

Then optimization kicks in, the meta re-forms, and the fun drains out again.

You’ve felt this loop. Maybe dozens of times. What I will show you today is that this loop is not about games.

It’s a fundamental property of any system where agents optimize against a shared landscape.

And it’s playing out right now in your career, your portfolio, your industry — at a speed that is increasingly inhuman.

Startups, Careers, and the Closing Window

The same dynamic is everywhere once you see it. But each domain reveals a different facet of the problem.

Startups show you the speed.

Early in a new market, the landscape is unmapped. Small teams with weird ideas can win because nobody knows what works yet. The first movers discover what the market rewards. Their strategies get copied. Playbooks get written. Best practices harden.

This used to take a decade.

Web 2.0 had years of green field. Remember going from AOL to Facebook? Feels like forever right?

Mobile had maybe five. I remember when the iOS App Store first came out, apps felt magical and fun. Now everything is either slop or corporate.

Web3 had 4 years, a compression I personally felt viscerally. One moment crypto was about decentralization and building the future of finance, then before you knew it the only way to make money was to buy attention and not be the last one holding the bag.

Generative AI had about eighteen months before the playbook crystallized. Raise a big round, hire a research team, build a wrapper or compete on infrastructure. The window between “anything is possible” and “the meta is locked” is compressing with every cycle.

Careers show you the trap.

When a new field opens up, the first entrants learn by doing. There are no credentials, no rubrics, no established path. Then the field matures. Degree programs appear. Hiring filters crystallize. The “right” background gets defined. The path narrows from “figure it out” to “follow the track.”

But here’s the part that connects to something deeper: the early entrants became good by exploring. They developed judgment by operating in an environment with slack, where there was room to make mistakes and learn from them. The people who follow the optimized track don’t get that training. They get the credential but not the capability. The system that produces experts stops producing experts once the meta forms.

This is the apprenticeship problem I wrote about in Agency Compression. The economy demands more cognition while removing the training ramps that produced it.

Markets show you the endgame.

Early markets have inefficiencies everywhere. Arbitrage is abundant. Generalists can compete. Then information flows faster, tools get better, the edges get competed away. The remaining alpha concentrates in the hands of whoever has the most compute, the best data, the fastest execution. Looking at you Citadel.

The market doesn’t become “unfair.” That framing misses the point. The market becomes solved. And solved markets don’t have room for new participants with conventional strategies. They have room for incumbents, and they have room for disruption. Nothing in between.

In every case, the same thing happens: optimization compresses the space of viable participation. Not by malice. Not by conspiracy. By the natural dynamics of agents competing for the same rewards on the same landscape.

The system gets better. The participants get squeezed.

The Squeeze

This tension — system performance vs. participant possibility — is the core of the essay. So let’s take a step back and make it concrete.

Imagine you’re a junior engineer in 2019. The meta hasn’t fully formed yet. There are half a dozen reasonable tech stacks, lots of companies hiring generalists, and the path from “I can code” to “I’m building something real” has a dozen viable routes. You can learn React and become a web developer, you can learn Rust and become insufferable, the world was your oyster. You can go to a startup or a big company. You can specialize early or stay broad. The space is wide enough that your particular combination of skills and interests can find a niche.

Now imagine you’re a junior engineer in 2026. AI has compressed the implementation layer. The baseline of “what a competent engineer produces” has been redefined by tools that didn’t exist two years ago. The hiring meta has shifted from “can you code” to “can you steer systems, hold context, and make decisions under ambiguity.” Except you were supposed to learn those things by coding. The ladder you were going to climb got optimized away while you were reaching for the first rung.

From the outside, the system looks healthier than ever. Productivity is up. Software is cheaper to produce. Companies are shipping faster.

From the inside, the possibility space just collapsed.

That’s the squeeze. The system improving and the participants getting compressed are not in tension. They’re the same process seen from different altitudes.

Why It Hurts

There’s a reason game designers ship resets. And it’s not because optimal play is “wrong.”

Think about the last time a game you loved was in a solved state. Not broken — solved. The tier list was settled. The builds were known. The “correct” way to play had been fully mapped.

Did you keep playing?

Most people don’t. Not because they’re bad at the game. Because the thing that made it feel alive — the possibility of discovery, the viability of your own weird approach, the sense that the outcome wasn’t predetermined — that thing is gone. The game is still running. You just stopped caring.

Now apply that feeling to a career. To a market. To an economy.

When possibility space collapses, something fundamental changes in the human experience. Work stops feeling like growth and starts feeling like a grind. Markets stop feeling like opportunity and start feeling rigged. The path from “where you are” to “somewhere meaningfully different” narrows until the only visible moves are incremental.

This isn’t a psychological footnote. It’s a core design constraint for any system that includes humans.

We need slack. Room to explore. Viable paths that aren’t fully mapped. The possibility that your weird approach might actually work. Not because we’re irrational. Because exploration is how we develop, how we find meaning, and how we produce the novel solutions that keep systems from stagnating.

A fully optimized system is a fully brittle system. It has converged on whatever it’s converged on. It has lost the capacity for search. It is maximally efficient at the current game and maximally fragile to any change in the rules.

The system needs resets too. It just can’t generate them on its own.

The Patch

A reset is any discontinuity that reopens possibility space that optimization had closed.

That’s it. That’s the definition. Not revolution. Not destruction. Not conspiracy. A reopening.

Resets exist on a spectrum. Small ones happen all the time: a new technology, a regulatory shift, a pandemic that reshuffles where and how people work. These are perturbations. They temporarily widen the space of viable strategies before optimization narrows it again.

Large resets are rarer: a new computing paradigm, a geopolitical realignment, a fundamental shift in what the economy rewards. These don’t just widen the space. They redraw the landscape entirely.

Game designers understand this intuitively. That’s why patches exist. That’s why seasons exist. Every competitive game with longevity has a deliberate mechanism for breaking and rebuilding the meta. The designers know that if they let the game converge, the players leave. So they inject disruption on a schedule.

But most real-world systems don’t have a game designer.

Markets don’t have a patch schedule. Careers don’t have a seasonal reset. The economy doesn’t ship balance updates.

Resets still happen. They just happen through crises, accidents, and creative destruction rather than through conscious design. And the faster the optimization cycle runs, the more frequently you need resets to keep the system healthy — but the harder they become to produce, because the optimized incumbents are getting better and better at defending the current meta.

This is the trap: the systems that most need resets are the ones most resistant to them.

The Acceleration

I wrote about Agency Compression — comfort without control, lives that get easier while becoming less authored. I wrote about the compounding loop — cheaper cognition compressing cycle times, pulling decisions upstream into machines.

This essay is the structural layer underneath both.

Agency compression happens because optimization collapses possibility space. The compounding loop accelerates because each cycle of optimization runs faster than the last. The need for resets is increasing at exactly the rate that our ability to produce them is decreasing.

The meta is forming faster than ever. The viable strategies are narrowing faster than ever. The cost of non-optimal participation is rising faster than ever. And the systems that benefit from the current arrangement are better than ever at defending it.

This is not a stable equilibrium. Systems that optimize without resetting don’t stay optimized. They stagnate, then they break. The question is never whether the reset comes. It’s whether it comes by design or by disaster.

The Deeper Thing

If you’ve been reading carefully, you may have noticed that every example in this essay — games, startups, careers, markets — is doing the same thing.

Optimization over a shared landscape converges. As the system approaches an optimum, the gradients flatten, the viable moves narrow, the system settles into a basin. Exploration yields diminishing returns. The system becomes increasingly committed to its current trajectory.

If you’ve spent time with machine learning, you know this shape. It’s gradient descent converging on a loss surface. And you know the solution: when the system gets stuck, you perturb it. Increase the learning rate. Inject noise. Warm restart. Shake the system out of its basin so it can explore new regions of the landscape.

Resets are warm restarts.

Games do this with patches. Markets do this through crises. Nature does this through forest fires and extinction events. Civilizations do this through revolutions and technological discontinuities.

The pattern is fractal. It recurs at every level of abstraction — from genetic propagation to market cycles to the rise and fall of empires. Optimization converges. Possibility collapses. A perturbation reopens the space. The cycle begins again.

This has been happening long before humans walked the earth. Thermodynamic systems, chemical equilibria, evolutionary dynamics — all of them exhibit the same rhythm of convergence and disruption. It’s not a feature of any particular system. It’s a feature of optimization itself.

And for the entirety of that history, the process has run blind.

The First Time

Cells don’t know they’re optimizing. Species don’t know they’re converging. Markets — despite being made of humans — don’t know they’re collapsing their own possibility space — otherwise crashes wouldn’t be so unpredictable. The resets come from outside: asteroids, ice ages, plagues, technological accidents. Creative destruction that no one designed.

The whole system, at every layer of the stack, has been running on autopilot. Optimize, converge, stagnate, break, reset, repeat. For billions of years. Across every domain. Without a single participant ever seeing the loop.

Until now.

Humans are the first participants in this process who can look at the meta forming and name it as meta-formation. Who can feel possibility space collapsing and understand why. Who can see the optimization converging and predict that the system is becoming brittle before it actually breaks.

This is not “this generation is special” rhetoric. This is a statement about the structure of complex adaptive systems. Self-awareness of optimization dynamics has never existed at any layer of the abstraction stack. Not in cells. Not in ecosystems. Not in economies. The process has always run without anyone watching.

We are the first node in the optimization graph that can see the graph.

The question — the real question underneath this entire essay — is whether that matters.

Because awareness doesn’t automatically produce action. You can understand Agency Compression perfectly and still get compressed. You can see the meta forming in real-time and still play the meta because the alternatives look unviable. You can read this essay, nod along, and change nothing.

Seeing the gradient is not the same as changing it.

But it’s a prerequisite.

No system has ever had the option of designing its own resets. Every reset in the history of complex systems has been imposed from outside — by chance, by catastrophe, by forces the system couldn’t see coming. For the first time, the possibility exists to do it on purpose. To build systems that preserve slack. To design institutions that resist full convergence. To engineer the equivalent of a patch cycle into markets, careers, and economies.

Not because it’s utopian, but because resets have a cost. The extinction of the dinosaurs made room for our ancestors, but if our ancestors went extinct then we wouldn’t be here. Too many resets too quickly can be detrimental to progress. We don’t need a big reset to make the meta feel fresh though, game designers already do it every season. The question is whether we can do it at the scale that matters.

The gradient does not point toward possibility space by default. It points toward convergence. Toward the meta. Toward a basin that gets harder to escape with every turn of the optimization cycle.

The meta is forming.

The patch isn’t scheduled.

But for the first time in the history of the universe, someone’s watching.

Personal plug time. If your engineering team is stuck in the local minimum of reading and writing code by hand, I can 2x your team’s productivity and raise the quality bar at the same time. Learn more about my consulting practice →

The same thing is about to happen with agents.

Just as the internet democratized access to information, AI agents will democratize access to digital labor. And just as most people were “on” the internet without capturing any of its value, most people will use agents without ever building one. The builders will eat.

Why Now

There is a fear right now that AI will do everything and humans will all get left behind. I do think this will happen someday, but the timeline will be much longer than most people think. True recursive self-improvement requires online learning. Until that happens, humans are stuck in the driver’s seat.

But this doesn’t mean that mass automation of knowledge work isn’t going to happen at scale. It absolutely is.

What changed is simple: for the first time in history, humans can amplify their unique knowledge and skills through a thing that can take action on their behalf. Not just generate text. Not just answer questions. Act.

In the past only business owners or government leaders had this power, and even then the control was fuzzy at best. You needed employees, org charts, communication overhead, alignment meetings. The friction was enormous. Now you can design and bring to life a digital being who is shaped by your unique taste, knowledge, and skills. This lets you scale yourself and create leverage.

Peter Steinberger showed us a version of this with OpenClaw. OpenClaw was not the first agent, but it was the one that woke up most of the non-developer community to what’s already possible. It demonstrated that anyone with their own expertise or niche can build a very lossy digital twin with the models we already have today. And this is the worst it will ever be.

Our digital twins are lossy right now because it doesn’t capture everything you know. It can’t replicate your full judgment, your intuition built from years of reps, your ability to read a room or a market. What it does capture is the codifiable part: your frameworks, your heuristics, your domain knowledge, your taste expressed as constraints. And that codifiable part, amplified by an LLM that can execute it 24/7 at near-zero marginal cost, is already enough to be dangerous.

The quality of that twin, how much signal it preserves versus how much it loses, is entirely a function of how well you design its harness.

The Harness

An agent at the end of the day is simply an LLM with some tools, acting on some environment. If you think of the agent as a robot, the LLM is the brain, and the harness is the physical body — what allows it to take action instead of just producing text.

Training a frontier-performance LLM is still prohibitively expensive for most use-cases. You’re not going to out-compete Anthropic or OpenAI on the model. So the key differentiator for most agents will be their harness.

A harness has three primitives:

Knowledge and skills. This is the agent’s unique expertise, delivered through things like skill files, memory systems, retrieval-augmented context, or curated document sets. This is where your taste lives. This is what makes your agent different from a generic wrapper around an API. If you don’t guard against slop or reward hacking, you will not get good results, because the harness is merely an amplifier for your taste.

Tools. These are the interfaces through which the agent interacts with the world: reading and writing files, calling APIs, browsing the web, executing code, sending messages. Tools define the action space. A model without tools is a brain in a jar. The tools are what turn inference into impact.

Triggers. Something has to cause the agent to act. It could be a request from the user. It could be a webhook, a cron job, a heartbeat that wakes the agent up periodically. Triggers define the agent’s relationship to time: whether it’s reactive or proactive, whether it waits or patrols.

These three primitives — knowledge, tools, triggers — are what determine the quality and capability of any agent. The model matters, but it’s increasingly commoditized. The harness is where the moat lives.

The Power Law Expands

Previously the power law in economics was reserved for individuals in fields with natural winner-take-all dynamics: sports, media, finance. The best captured a disproportionate share of the value because their output was uniquely theirs and infinitely distributable.

Agents extend that dynamic to almost every field.

You can create the best coding agent and make it available to everyone. You can build an agent that embodies your 20 years of supply chain expertise and sell access to it on a marketplace. You can let it get “hired” like an employee would, running tasks for clients while you sleep. Your knowledge becomes a product with near-zero marginal cost of delivery.

This is what makes agents different from SaaS, from consulting, from courses. Those were all attempts to scale expertise, but they all hit human bottlenecks: you had to write the content, take the call, maintain the product. An agent scales the expertise itself. It acts on your behalf, applying your judgment to novel situations, compounding your leverage every hour of every day.

The people who figure out how to encode their unique knowledge into high-quality harnesses will capture a disproportionate share of the value being created. Everyone else will use their agents.

The Compounding Loop

This isn’t a one-time shift. It’s a flywheel.

Better harnesses produce better agent output. Better output attracts more users and use-cases. More use-cases generate more data about what works. More data improves the harness. The harness gets better.

And there’s a second loop running underneath: as agents get deployed more widely, the tooling and infrastructure around agent development improves. Better frameworks, better evaluation harnesses, better patterns for knowledge encoding, better marketplaces for distribution. This lowers the barrier for the next wave of builders, which expands the surface area of what agents can do, which pulls more economic activity into the agentic layer.

This is the same compounding loop I wrote about in my predictions on the future of work. Cheaper cognition expands the economic surface area, compresses cycle times, and rewards fast actors, which pulls more decisions upstream into agents and systems, which makes cognition even cheaper. The flywheel never sleeps.

The difference is that right now, in the early phase, humans are still the ones building and shaping these agents. The harness is still a human artifact. Your taste, your knowledge, your judgment, these are the inputs that determine whether an agent produces value or noise.

That window won’t stay open forever. But it’s wide open right now.

So What

I predict that the key differentiator between agents will be these three harness primitives over the next 1–2 years. If you are looking to become one of the most valuable humans in the economy, you should be studying agent design right now.

Even if you don’t write your own agent, even trying to add your unique knowledge to an existing agent requires you to understand how that agent works, what its knowledge layer looks like, what tools it has access to, what triggers drive its behavior. You need model empathy: an understanding of how these systems think, where they break, and how to shape their output through constraints rather than hoping for the best.

The people who build the best websites in 1998 didn’t just know HTML. They understood distribution, user behavior, and how to turn attention into value. The people who build the best agents won’t just know prompt engineering. They’ll understand harness design, taste encoding, and how to turn expertise into leverage.

The gradient does not point toward us by default. But right now, the tools to shape it are sitting on the table.

If you or your engineering team want to get ahead of the agentic revolution rather than react to it, I help teams and individuals design high-leverage agent systems and re-orient their engineering practice for an AI-native world. Learn more about my consulting practice →

That question has many proponents from multiple sides, from Universal High Income, to the extinction of humanity. However today I’m looking to address a different question, one that I think is easier to reason about and more relevant to what we can do today to make a difference.

The real question is whether humans will remain economically relevant as actors or whether we become a small legacy market, akin to an amish community operating next to a big metropolis.

History is a sequence of shifts in what the economy rewards, and this time humans might have nothing left to offer.

History of Disruption

The shift from hunter-gatherer to agriculture forced humans to accumulate different skills: storage, planning, settlement, social structure. The change was slow and multi-generational, culture had time to recompile

The shift from small cities to large urban centers rewarded specialization. Again this was a slow change, the economy got more complex but evolved at human generational speeds.

Industrialization compressed the timeline further. People moved into cities for factory work, within a few decades, a lifetime. Factory work was grueling, but there was very little skill required at least for entry level jobs (by necessity) and so people managed to adapt.

Then the internet happened. Disruption started to feel too fast, but it was contained mostly to new industries and skillsets that were created. A few jobs disappeared and a few industries were restructured but it was largely additive in nature. Most people felt the internet adding new ways to do their job instead of replacing.

Now with the AI revolution people are saying the same thing will happen, that the pattern will continue. But this time is different. AI is a fundamental shift not in the way that we do things, but in the things that we can do which are better than AI. Each day that pool of tasks shrinks.

Additionally, each revolution stacks on the previous one. AI arrives after we’ve already built a global nervous system. It has the internet as a distribution mechanism. Change only gets faster, so expect this not to take decades, but years. Humans just weren’t evolved to design with changes happening this fast.

That’s why I think this wave may finally test the limits of human adaptability. This doesn’t mean all jobs disappear overnight, rather we will enter a period of gradual disempowerment: humans lose leverage as value creators and as decision makers. You can stay employed and still become irrelevant. You can get richer in absolute terms and still shrink as a fraction of the economy.

And the most dangerous part is that the transition will come wrapped in prosperity.

The False Plateau

In the early phase, AI will feel like an economic miracle.

Software becomes abundant. Digital content becomes hyper-personalized and hyper-addictive. Creation costs collapse. One person can do what used to require a team. Every business gets a productivity jump. Individuals feel like they’ve gained superpowers.

This is the false plateau. Because while it looks like empowerment, what’s really happening is a quiet relocation of agency. The center of gravity moves away from humans as authors of decisions and toward humans as inputs to systems that decide. Even before generative AI this was already happening.

Even today, it’s not obvious that the CEO of a large company is truly at the helm. Not because CEOs are incapable. Because the system they operate in has become too complex, too fast, and too information-dense for any individual human to fully hold.

Decisions don’t feel like authored choices anymore. They feel like gradients.

Markets apply pressure. Metrics become reality. Incentives propagate. A thousand micro-decisions get made “because the dashboard says so,” “because the competitor shipped it,” “because the model predicts it,” “because the quarterly cycle demands it.”

It’s like messenger proteins in a cell. Signals bind to receptors. Cascades trigger cascades. The super-organism moves.

AI accelerates this because it makes signal-processing cheaper and reaction-time shorter. Once you can cheaply generate, summarize, decide, negotiate, monitor, and iterate, coordination becomes machine-native. And coordination is the control plane. It’s where the leverage is. Jobs are downstream of that.

The Compounding Loop

From first principles, the core dynamic looks like a flywheel:

1. Cheaper cognition and coordination expands the economic surface area (similar to how cheaper compute made more software viable)

2. That expansion compresses cycle times and increases complexity (similar to what the internet did for information)

3. Compressed cycles reward fast actors, machines can act at light speed and in parallel, 24/7 (similar to how trading firms became fully digitized over time)

4. That pulls more decisions upstream into agents and systems

5. Which makes cognition and coordination even cheaper, turning the flywheel once more

This is why the transition is jagged. But the waves get faster and faster. Less cycle time in-between. And with each turn of the wheel, more humans are tossed out of the loop.

Phase 1: Skill Bifurcation

At first it will look like a skill issue. People who can leverage AI become dramatically more productive. Many white-collar jobs get unbundled into workflows. Whole teams become a single person monitoring teams of agents. Entire job functions become cheap: writing, coding, research synthesis, design drafts, sales collateral, data analysis, the list goes on.

Meanwhile the economy still needs humans for a lot: oversight, services, embodied trust, physical labor, field operations, etc… so it doesn’t look like humans are obsolete, it looks like every other transition before it. People get sorted into new, hopefully better and less mundane jobs. This is where most “adapt and you’ll be ok” narratives end.

But the flywheel keeps turning.

Phase 2: Ownership Bifurcation

Then it becomes an ownership issue. The moat is not “knowing how to use AI” the moat is owning resources like models, data, distribution, compute, energy, aka capital in all forms.

This is when the false plateau becomes dangerous. People will feel richer than ever, the economy will do better than it ever has, all while the control plane quietly consolidates. And the consolidation will be sticky, because coordination advantages compound.

And the flywheel keeps turning.

Phase 3: Capital Bifurcation

Eventually even wealthy humans will start being displaced. This sounds outlandish, but it’s a simple extension of the coordination argument. Wealth is leverage only if it can be converted into timely, correct decisions at the frontier. If decision velocity and complexity keeps increasing, then un-augmented human governance becomes the bottleneck. Not because humans are dumb, but because humans are slow, serial, bad at consensus building, and expensive compared to systems that can think in parallel, operate continuously, and update globally.

Slowly, the first non-humans will start to “own” things. Human ownership without competitive steering becomes ornamental. The system routes around bottlenecks. The world doesn’t end. It just stops being human-centered. [If you want a hard-scifi take on this path, give “Accelerando” by Charles Stross a read].

The flywheel never sleeps.

The Alien Economy

Over time, humans shrink as a fraction of the economy as autonomous AI systems start serving other AI systems. Once machine agents become meaningful customers of other machine agents, human taste becomes less and less relevant to profitability.

The objective functions increasingly become set by emergent equilibria: financial markets, information battlegrounds, competitive dynamics between agents, resource allocation games beyond our comprehension. Humans don’t have to be hated or harmed in the process, humans just have to be uncompetitive.

The strange part is that humans will keep getting richer in nominal terms, but the economy will grow far beyond what humanity controls, and humans will become an inconsequential corner of the global super-organism.

So What

This is a work of speculation. But it’s grounded on real trends that have already been shaping humanity for thousands of years. It’s a warning that the ship we’re on is really hard to steer, and that there’s a big ice-burg right in front of us.

However I’m a realist and a technologist. There exists no “stable state” utopia solution. The dynamics of the world are always changing. The only constant is change. Therefore I don’t think that “halting AI progress” or techno-deceleration is a realistic nor feasible approach. I don’t even think that AI alignment is about keeping humans in the loop forever.

I think alignment is about something more pragmatic and more urgent: Creating an uplift path for humanity so we can continue to shape and experience our own light cone.

In other words, alignment is a bridge, not a destination. The main race isn’t “humans vs AI tools.” The main race is AGI → ASI, where tools become actors. The risk is not that systems get smart. It’s that autonomy and optimization pressure outpace human steering ability before we’ve built a viable way to stay in the game.

So the goal isn’t to freeze the economy in a human-centered equilibrium. The goal is to preserve enough human control, leverage, and option value to cross the threshold into a world where “human vs AI” stops being a meaningful distinction.

I am only human after all. I would like to believe that there’s something special about us that’s worth persisting. Worth carrying into the stars. I want to make sure that whatever being inherits the galaxy has an innate curiosity, empathy, and optimism that humanity embodies. I want the thread of humanity to run through our decedents, not just as a footnote or as a biological boot-loader, but as the soul of future machines.

The gradient does not point toward us by default. And if we wait until it’s obvious, it will already be too late to matter.

Subscribe now