The Last Gate

On the dating of skills, the dissolution of gatekeeping, and why energy is the final frontier — at the dawn of a new imperium and the slow failure of the old one

Every imperium fails on the same two things: communication and efficiency. Rome did not fall because of barbarians. Rome fell because the cost of moving information and goods across its territory exceeded the value extracted from holding the territory. The barbarians arrived afterward, the way scavengers arrive at a body. The Spanish empire did not collapse because of military defeat. It collapsed because silver from the Americas debased its own currency faster than its administrative capacity could adapt, and the gap between what the empire was trying to coordinate and what it could afford to coordinate became larger than the empire could survive. The British empire didn't end because the colonies revolted, exactly. It ended because the energy cost of running the global system stopped being recoverable from the system itself, and the empire quietly liquidated.

This is the pattern. Every large coordinated human system reaches a point where the maintenance cost of coordination exceeds the value coordination produces, and at that point the system either reorganizes or disintegrates. The reorganization is sometimes fast, sometimes slow, sometimes successful, sometimes catastrophic. The pattern doesn't care about the specifics. It cares about the ratio between what you're trying to hold together and what it costs you to hold it together.

We are at one of these points. We are at it for all of our coordinated systems simultaneously, which is novel. And the technology that's about to reshape this transition is being built on top of the same exhausted substrate that's failing underneath it. This is the situation. Most of what's written about AI right now misses this because it focuses on the surface — what the models can do, who benefits, what jobs change. The surface is real but it's downstream. The structural story is that gatekeeping is dissolving, the bottleneck is moving to substrate, and the final frontier — the one nobody escapes — is energy.

The dating of skills

There used to be a craft called computing, performed by humans who did arithmetic by hand for astronomers and ballistics engineers. The skill was real, the practitioners were skilled, and the work mattered. Then the mechanical calculator dated the skill. Then the electronic computer dated it again. By the time anyone alive today was born, "computing" as a human skill was an artifact of history. The people who held that skill weren't bad at it when the dating happened. The dating wasn't about quality. It was about the ground moving underneath the category itself.

This has happened many times. Cartography was a high craft until satellite imagery dated it. Typesetting was a guild trade until desktop publishing dated it. Translation was a multilingual elite's domain until machine translation dated most of it. Each time, the skill that was being dated felt impossible to date from the inside — what do you mean a machine could draw a map, what do you mean a machine could set type, what do you mean a machine could translate Russian. Each time the answer turned out to be "yes, and the skill becomes a hobby or a specialization for edge cases, while the bulk of the work moves underneath the new system."

The dating is never the part that hurts. The part that hurts is that the dating is invisible until it's already happened. The cartographers in 1985 did not know they had ten more years of their craft being primary. The translators in 2005 did not know what 2020 would look like. The radiologists in 2015 still don't quite know what 2030 will look like, and they argue about it instead of preparing for it, which is the normal human response to being inside a category that's being dated.

What's different now is the scope and speed of dating. Previous waves of dating moved slowly enough that careers could complete inside them. Cartographers who started in 1960 could retire in 2000 still doing meaningful work, even if the field had narrowed. The current wave is going to date entire categories of cognitive labor — programming, writing, design, analysis, much of medicine, much of law, much of management — inside a single career, possibly inside a decade. The people inside those categories have not absorbed this yet, and most of them will not absorb it until it's too late to position for the next thing, because the dating is happening from underneath and the surface still looks normal.

The first frame to understand is this: every gate that selects for cognitive skill is currently being dated. The gates that selected for programming ability, for writing ability, for analytical ability, for the ability to hold complex systems in your head — these gates were real and they were doing real work, but the substrate that made the work necessary is dissolving. The skills that cleared those gates are becoming the equivalent of beautiful handwriting after the invention of the printing press: still a thing, still admirable, no longer load-bearing.

Gatekeeping moves, it doesn't disappear

The naive reading of the dating-of-skills story is that AI flattens the field and everyone can now participate equally in cognitive work. This reading is wrong, and it's wrong in a specific way that matters.

Gatekeeping doesn't disappear. It moves. Every time a gate is dissolved by technology, a new gate forms upstream of it, selecting for whatever the new technology can't do yet. The dissolution of the syntax-and-execution gate in programming didn't create a flat field of equally capable software builders. It revealed that the real gate was always upstream — at the level of structural thinking, problem decomposition, taste. People who'd been clearing the visible gate (knows the language, can write working code) suddenly discovered there was a less visible gate they hadn't been clearing, and the people who had been clearing the upstream gate all along now had radically more leverage because the downstream tax had dropped to near zero.

This pattern is general. Every dating of skills produces a brief egalitarian moment, then reveals a new tier of selection happening at a higher level of abstraction. The people who win the new regime are not always the people who won the old one — sometimes the cognitive style required is different enough that there's a partial reshuffling — but the regime itself is no flatter than what it replaced. It's just selecting on different attributes.

What this means concretely: the current AI moment looks democratizing because the visible tax on cognitive work is dropping. Anyone can produce code, write articles, generate images, analyze data. This is real and it is partially democratizing. But it is also revealing latent capability gaps that were previously hidden under the uniform execution tax everyone paid. The people who can articulate problems clearly, hold structure in their heads, recognize good output from slop, and direct AI systems toward worthwhile ends are scaling at a rate that's incommensurate with anyone who can't do these things. The middle is hollowing out. Not symmetrically — the bottom is rising slightly in absolute capability while falling relatively, and the top is going vertical.

This will continue until the next gate dissolves, which will happen as soon as AI systems become capable enough to also do the articulation, the structural thinking, the taste-making. At that point the next upstream gate becomes visible, and so on. The dissolution proceeds, the gates keep moving up the cognitive ladder, until eventually they reach the point where the ladder ends — where the constraint is no longer cognitive at all.

The end of the cognitive ladder

The cognitive ladder has a top. The top is harder to see clearly from where we currently stand, but its shape is becoming visible.

Past a certain capability level — call it weak AGI, call it whatever — AI systems become capable of supplying their own structure. They can articulate problems better than the users asking them. They can predict intent before the user has finished forming it. They can generate options, evaluate them against learned models of user preference, and converge on good answers faster than the user could specify what "good" means. At this point the user-side cognitive gate doesn't just thin. It dissolves. The skills that selected for cognitive merit in the previous regime stop being load-bearing for any economic outcome, because the AI is doing those things too.

When the cognitive gates fully dissolve, the constraint moves off the human side entirely and onto the substrate. What remains as the actual bottleneck is:

• Compute. The physical capacity to run inference at the scale required for intent prediction and intelligent action. This is constrained by chip fabrication, by datacenter construction, by the supply chains feeding both.
• Energy. The electricity required to power compute. This is the most fundamental constraint of all, and it's the one that turns out to be the final frontier. Every other constraint can in principle be relaxed by investment, engineering, or time. Energy is the constraint that doesn't relax — it has thermodynamic floors, it has political and geographic distribution problems, and the rate of demand growth is now exceeding the rate of supply growth in most developed economies.
• Coordination. The ability of human institutions to deploy, govern, and benefit from AI systems faster than the systems destabilize the institutions. This is currently failing visibly in real time, and the failure is not unique to AI — it's the same coordination failure showing up in climate response, in pandemic response, in infrastructure maintenance, in democratic governance. AI is making the coordination problem harder, not easier, in the near term.

These are the actual gates of the next era. They are not cognitive gates. They are not gates that can be cleared by being smart, working hard, or articulating well. They are infrastructural and political gates, and they're being seized by the parties best positioned to seize them — frontier AI labs, major cloud providers, large states with sovereign compute capability — while the rest of the economy looks the other way.

This is the part the productivity-tool framing of AI completely misses. The story is not "AI helps you write code faster." The story is "AI is the surface on which a much deeper restructuring is happening, where the constraint of the economy moves from human cognitive labor to energy and compute, and ownership of that constraint becomes the new form of power."

Energy as the final frontier

Every previous transition between economic regimes can be told as a story about energy. The agricultural revolution was solar energy captured through plants and managed through human and animal labor. The industrial revolution was fossil energy unlocked at scale, multiplying human productivity by orders of magnitude. The information revolution was electrical energy applied to computation, creating new categories of value that didn't exist in the previous regime.

The AI transition is the same story at a higher level. AI systems are vast energy-to-cognition transformers. Training a frontier model requires the energy output equivalent to a small city for months. Inference at scale requires datacenters whose power requirements are starting to outpace the grids they're built on. The marginal cost of an additional unit of useful AI output is, fundamentally, a marginal cost of energy. Every other input is amortized; every prompt is electricity.

This is starting to show up in places you might not expect. Major AI labs are building or commissioning their own power generation — including nuclear — because they cannot rely on grids that were designed for an economy with different demand curves. Cloud providers are siting datacenters near power sources rather than near customers, because the new economic geography is dictated by electricity rather than by users. Nation-states are evaluating compute capacity the way they evaluated industrial capacity in the twentieth century, because compute is industrial capacity now.

The countries that own the energy own the next regime. The countries that can build power generation faster than their AI demand outpaces it remain players. The countries that can't will become tributary economies to the ones that can, regardless of how clever their populations are or how good their universities are. Cognitive merit at the national scale becomes irrelevant once the energy ceiling is binding. You can have all the brilliant engineers you want. If they're running on imported compute powered by imported electricity, you're not running the regime — you're a customer of it.

This is why the geopolitical maneuvering around chip manufacturing, around power generation, around datacenter siting, around export controls on advanced semiconductors, is the actual story of the next decade and not a sideshow. It looks technical and bureaucratic because it is technical and bureaucratic. The thing being fought over is who controls the substrate of the next economic regime. The fight is happening now. Most of the people fighting it are not communicating about it publicly, because they don't need to.

The dawn of a new imperium

Imperial transitions look chaotic from inside because they are chaotic. The Roman transition produced four centuries of disorder before the rough shape of medieval Europe stabilized. The transition from feudalism to early modernity produced a century of religious wars, civil wars, and revolutions. The transition from the European imperial order to the postwar order took two world wars and a depression. Each transition was navigated by people who mostly didn't understand what was happening at the time, and who were operating with mental models calibrated to the previous regime.

We are at the front edge of a transition of the same scale. The previous regime — call it the post-1945 order of nation-states with market economies coordinated through international institutions — is exhausted by every measure that mattered to it. The institutions designed to manage it are visibly failing to address any of the major challenges of the present, from climate to migration to economic inequality to information disorder. The political coalitions that supported it have come apart in nearly every major democracy. The international system it was supposed to coordinate is fragmenting into rival blocs. The signs are not subtle.

What's coming next is not yet clear. But the shape of it is starting to be visible, and the shape has several features:

The new imperium will be organized around control of energy and compute rather than around control of territory. Territory still matters, but territory without energy is increasingly worthless, and energy without sovereign compute is increasingly tributary. The states that figure this out and act on it will be the great powers of the next era. The states that don't will be raw material suppliers, regardless of their current GDP rankings.

The new imperium will be more concentrated than the previous one. The post-1945 order, for all its flaws, included a relatively wide distribution of meaningful sovereignty among middle-sized states. The compute-and-energy regime concentrates power among a much smaller number of actors, possibly fewer than ten globally. The "rules-based international order" was a system of negotiation among many parties. The new order will be a system of negotiation among very few parties, with everyone else accommodating their decisions.

The new imperium will be intermediated by AI in ways we don't have language for yet. The institutions of the previous regime — parliaments, courts, ministries, central banks, international bodies — were human-staffed bureaucracies coordinated by paper and email. The institutions of the next regime will be hybrid human-AI systems that operate at speeds and scales no purely human institution can match. Whoever builds these institutions first will set the patterns the rest of the world has to either adopt or resist.

The transition between regimes will be painful in the specific ways imperial transitions are always painful. Coordination will fail. Trust will fail. Systems people depended on will stop working. People will lose savings, status, relationships, sometimes lives, to changes they didn't see coming and couldn't have prepared for. The institutions that should have managed the transition will mostly fail at it, because they were designed for the previous regime and can't adapt fast enough. Some of those failures will be catastrophic. Most of them will be ordinary — slow degradation of services, quiet collapse of expectations, accumulating small losses that add up to a different world.

Every system is failing on the same things

The thing that should be most disturbing about the current moment is not any single failure. It's that all the failures rhyme. The same root causes — communication breakdown and efficiency exhaustion — are showing up in completely different domains simultaneously.

Healthcare systems in developed economies are failing because the cost of coordinating care exceeds what the systems can fund, while administrative overhead grows faster than clinical capacity. Education systems are failing because the cost of producing the credentials they sell exceeds the value of the credentials in the labor market, while the systems can't adapt to changing demand. Democratic institutions are failing because the cost of coordinating consent across fractured information environments exceeds what the institutions can produce, while polarization makes coordination harder. Cloud infrastructure is failing because the cost of securing complex distributed systems against AI-augmented attackers exceeds what the providers can invest, while user expectations of reliability rise.

These are the same failure. Different surfaces, same underlying dynamic. Coordinated systems hitting their efficiency walls, with no obvious path to reorganization that preserves the value the systems were producing.

The historical pattern when this happens is not that the systems get fixed. The pattern is that some of the systems get replaced, often violently and chaotically, and the others linger as zombie institutions for decades, performing rituals of their previous function while not actually accomplishing what they were designed to accomplish. Western Europe in the seventh century had Roman roads, Roman law, Roman administrative categories — all increasingly unrelated to what was actually happening on the ground. Late Qing China had imperial examinations producing bureaucrats for a state that no longer functioned. Late Soviet Russia had five-year plans being filed by ministries that everyone knew were lying.

The zombie phase is the long one. The replacement phase is the short one. We are in the early zombie phase across most of our major institutions, and we don't yet know what replaces them, or whether anything does, or whether the replacement is worse than what came before. Historical precedent suggests it is roughly even odds in either direction.

Where this leaves an individual

There is a temptation, looking at all of this, to either despair or to try to position oneself as a winner of the new regime. Both are mistakes.

Despair is a mistake because the actual transition will be long, uneven, and full of pockets of normal life. Most people in seventh-century France were not in active crisis at any given moment, even though they were living through what historians call the collapse of Rome. They were farming, raising children, gossiping, having ordinary days. The collapse was visible in the gaps — in the things that used to work and no longer did, in the slow loss of capacities that nobody could recover — but it was not a continuous emergency. People who treated it as a continuous emergency were exhausting themselves uselessly. People who treated it as a long shift in the conditions of life adapted, slowly, to the new conditions.

Trying to position as a winner is a mistake because almost nobody knows what the winning positions are, and the few who do can't be displaced by ordinary individual effort. The positions that will matter — control of energy infrastructure, of compute, of foundational AI capability, of the political mechanisms that govern these — are not available to most people, and pursuing them as an individual is mostly a way to be exploited by the actors who already hold them.

The honest individual response to imperial transitions is roughly the same across history. Reduce your exposure to the failing systems where you can. Maintain skills and relationships that don't depend on the specific arrangements of the failing regime. Don't trust institutions whose interests don't align with yours, regardless of how respectable they appear. Keep a margin of resilience for the failures that will reach you personally, because some of them will. Be skeptical of anyone selling you a clear path through the transition, because no such path exists. Live a life that would still be worth living if the larger systems failed harder than expected, because the value of a life has never primarily come from the systems it's embedded in.

This is the wisdom of every era of transition. It is boring, it is hard to monetize, and it does not generate engagement. It is also approximately true.

And, in passing, a note on what's coming in the near term

Before closing out, one practical note that follows from the larger analysis but doesn't need the analysis to be useful.

The cloud-mediated digital life that most people in developed economies now depend on is a specific arrangement that emerged in the last twenty years and is, by the standards of this essay, late in its phase. The trust model — that you can put your photos, documents, communications, and identity into the systems of a few large providers and they will reliably remain accessible to you — is now showing the same kinds of stress that other late-regime arrangements show. Mass account terminations driven by automated systems. Data losses with no recovery. Customer support that doesn't exist for anyone below enterprise tier. Moderation errors at scale with no appeal. Increasing attack surface as AI capabilities accelerate on both offensive and defensive sides.

Inside the window we're entering — call it the next two years, give or take — at least one large-scale incident affecting a substantial population of end users is highly likely. The shape of it could vary. A million accounts incorrectly terminated by an AI moderation system. A successful attack on cloud identity infrastructure that locks people out of services they depend on. A major data corruption event that destroys photos and documents across a provider's user base. The specifics matter less than the pattern: the trust model is fragile, the defensive capacity has not kept pace with offensive capability, and the incident that exposes this is approaching.

The response to this, if you don't want to be among the people who find out the hard way, is straightforward and entirely within your control. Have at least one copy of anything you care about that is not on a phone, not in the cloud, and not dependent on any account or service that could be terminated or compromised. A physical drive that you own, stored somewhere safe, updated occasionally. That's it. That's the whole intervention. It is the equivalent of keeping some flour and oil in the pantry: not paranoid, not preparation for the end of the world, just an acknowledgment that the supply chains you depend on are not infinitely reliable and a small margin of self-sufficiency is sane.

This is not the most important thing in this essay. The imperial transition, the energy frontier, the dissolution of cognitive gates — those are the larger story. But the local backup is the one piece of action available to most readers, and refusing to mention it because it sounds quotidian would be a kind of cowardice. People who don't do this will, in non-trivial numbers, lose things they cared about over the next several years. The drive in the drawer is the simplest possible insurance against being one of them.

Closing

We are at the dawn of a new imperium, and like every imperium before it, the old one is failing on communication and efficiency, the same way they all do. The cognitive labor that organized the previous economy is being dated by AI faster than the institutions built around it can adapt. The gatekeeping that selected the previous era's winners is moving upstream, toward taste and judgment, and then dissolving entirely as AI capability catches up. What remains as the final frontier is the substrate itself — energy, compute, and the political control of both.

This is the actual story. Most of what's written about AI right now is froth on the surface of this story. The productivity gains are real but downstream. The job displacement is real but partial. The capability surprises will keep coming, and most of them will be subsumed into the larger pattern within months of their arrival.

The new imperium will be organized around control of the substrate. The actors building it know this. The institutions of the previous regime mostly don't, or can't say so publicly. The transition between regimes will be uneven and painful in ordinary as well as dramatic ways, and the wisdom for navigating it is the same wisdom that has applied to every previous imperial transition: reduce exposure to the failing systems, maintain capacities that don't depend on them, don't trust institutions that have stopped serving their stated purposes, and live a life that doesn't require everything to keep working as it currently does.

The wild years are starting. They will be wild for a while. On the other side of them is a different world, organized differently, with different gates and different winners and different losers and a different texture entirely. We don't know which one yet. The honest position is to acknowledge the uncertainty, prepare for the parts you can prepare for, and pay attention without becoming captured by the attention economy that profits from your alarm.

And — by the way — back up your photos to a drive. That part is easy.

The patterns of imperial collapse and renewal are old. The substrate is new. The combination is what makes this moment historically distinctive and personally consequential. Reasonable people can disagree about timing and details. The shape is not seriously in dispute among people who have looked at it carefully.