In 28 BC, Octavian declared the emergency powers of the civil-war era null and void. The following January, he rose from his seat in the Senate house. Since the age of nineteen, when he spent his own fortune to raise legions, he had been the man who ended a century of civil war. After crushing Antony and Cleopatra at the Battle of Actium, all remaining military force in Rome was concentrated in his hands. What he said before the Senate was brief.
"I return all power to the Senate."
The Senate reciprocated. They bestowed upon him the title Augustus — "the revered one" — and entrusted him with responsibility for the "troubled provinces" for ten years. The fact that most of Rome's legions were stationed in those "troubled provinces" was something the Senate knew, and Augustus knew. Whether the renunciation of emergency powers in 28 BC and the "return" of authority in 27 BC were a single choreographed performance or two distinct events remains a matter of scholarly debate.1 Either way, the result was the same — one-man rule began under the forms of the Republic.
The forms of the Republic were preserved. The Senate endured. Voting, debate, the election of consuls — all continued. But real power flowed to one man. This was the birth of the most sophisticated fast order history has recorded.
In Chapter 1, we saw the Capitoline Hill. The year 133 BC, when Tiberius Gracchus stumbled and fell as he fled. The hill where senators beat a fellow citizen to death with chair legs. What Gracchus had attempted was to enforce a law that had gone unenforced for 234 years. It was an attempt at slow justice. It was crushed. A hundred years of civil war followed, and at the end of that century stood Augustus.
When slow justice was crushed, fast order came.
And fast order worked. The Pax Romana — roughly 207 years of relative peace, from 27 BC to the death of Marcus Aurelius in AD 180.2 Roads, aqueducts, harbors, legal codes. Augustus boasted that he "found Rome a city of brick and left it a city of marble." He was not wrong.
But beneath the marble, something was vanishing. The fierce political oratory that had echoed through the Forum in Gracchus's day vanished first. The forms of Senate debate survived, but substantive opposition was no longer safe. Caligula and Nero rose atop that structure. The succession problem was never solved, and the system degenerated in the Crisis of the Third Century.
At the end of Chapter 11, we asked: Was fast order truly good order?
To answer that question, we turn to history.
1. Three Faces of Fast Order
At every moment of crisis, history has produced figures and systems that propose fast order. They share one thing in common. They gain speed by bypassing consensus. The price only becomes visible with time.
Augustus. From 27 BC to AD 14. He executed the transition from Republic to Empire not as revolution but as gradual institutional transformation. The direction was consistently toward one-man rule, but the forms remained republican. The achievements were real — 207 years of Pax Romana, provincial administrative reform, centralization of the tax system, decentralized integration. The costs were equally real — the hollowing-out of the Senate, the decline of civic political participation, unresolved succession crises, and the loss of self-correcting capacity. When honest information flow from bottom to top was blocked, the institutional pathways for correcting bad policy disappeared. Augustus himself was capable. But the system had no mechanism for guaranteeing capable successors.
The Napoleonic Code. 1804. After the French Revolution destroyed the ancien régime but failed to build a stable alternative, Napoleon promised order to a France exhausted by chaos. A committee of four jurists drafted the code in four months. The Conseil d'État held 107 sessions; Napoleon personally attended 55 of them, reviewing 2,281 articles with the pragmatism of a soldier who had never studied law.3 Thirty-six individual laws (lois) were passed between March 1803 and March 1804, then consolidated and promulgated as a single code on March 21, 1804. In exile on Saint Helena, he said: "My true glory is not the forty battles I won. What will live forever is my Civil Code."4
He was right. More than 220 years later, the Code continues to shape the law of over forty countries. It abolished feudal law, established equality before the law, and guaranteed property rights. Yet the same Code erased women's rights that had been partially recognized during the Revolution, and restored colonial slavery that the revolutionary government had abolished in 1794. And the regime that produced the Code itself collapsed in eleven years. Fast order can produce great institutions, but it cannot guarantee the survival of the system that sustains them.
China's Reform and Opening Up. December 1978. The strategy of "crossing the river by feeling the stones" — attributed to Deng Xiaoping, though sometimes to Chen Yun. Not wholesale liberalization but special economic zone experiments; expand nationwide if they succeed. The speed of a party-state that could designate special economic zones, attract foreign capital, and relax labor regulations without parliamentary approval made it all possible. Shenzhen in 1980 was a fishing village of roughly 30,000 people (about 330,000 in the broader administrative district).5 Within a decade, Shenzhen's GDP had grown sixty-fold and industrial output two hundred-fold — a transformation that earned the phrase "Shenzhen speed" (深圳速度). China's overall GDP grew from $150 billion to $18.74 trillion.6 The absolute poverty rate fell from 41 percent to below 5 percent.7
It was the fastest economic growth in human history. The costs were equally historic — Tiananmen in 1989, internet censorship, Uyghur surveillance, press control, and Xi Jinping's abolition of the ten-year term convention Deng had introduced. Augustus's structural vulnerability repeated itself: one-man rule without institutionalized succession, and a system in which information from below never reaches the top.
The common pattern across all three cases is clear.
The mechanism by which fast order gains speed: it bypasses consensus. It bypasses the gathering of information. It suppresses dissent. It conceals the costs of trial and error.
The result: short-term stability, long-term fragility.
2. Three Faces of Slow Justice
Slow justice operates in the opposite direction.
The British Parliament: 240 years. From the Glorious Revolution of 1688 to universal suffrage in 1928. It took 240 years. The First Reform Act of 1832 extended the vote to roughly 7 percent of the total population — one in seven adult men — and only to middle-class men who met certain property requirements.8 Not until 1867 were urban workers included, doubling the electorate to approximately two million. In 1884, rural laborers were added. In 1918, women over thirty gained the vote. Only in 1928 could all adults aged twenty-one and over finally cast a ballot.
Two hundred and forty years. It looks like inefficiency at its most extreme. Yet during those 240 years, Britain achieved democracy without revolution. When revolutions erupted in chain reaction across continental Europe in 1848, Britain remained stable — the Chartist movement submitted a petition to Parliament claiming 5.7 million signatures (Parliament's own verification found roughly two million), but the outcome was "petition instead of revolution."9 During the same period in which France cycled through revolution, empire, restoration, republic, empire, and republic again, Britain accomplished fundamental change within the same institutional framework. It survived because it was slow.
The sixty-four years of factory legislation we saw in Chapter 2 — from Arkwright's spinning frame in 1769 to the factory inspectors of 1833 — were a microcosm of this British system. Laws were made, they failed, and better laws were made. The cycle of "law → failure → better law." Slow, but a slowness that learned.
The United States Constitution: 27 amendments in 247 years. Deliberately designed to make amendment difficult — a two-thirds majority in both the House and the Senate, ratification by three-fourths of state legislatures. The most recent amendment (the Twenty-Seventh, on congressional pay) was proposed on September 25, 1789, and not ratified until May 7, 1992 — 203 years.10 Had University of Texas undergraduate Gregory Watson not argued in a 1982 term paper that it was still eligible for ratification, the amendment might have slept forever. Abolishing slavery took 89 years (1776–1865). Women's suffrage took 144 years (1776–1920). During those 89 years, during those 144 years, real human beings lived in suffering.
We must acknowledge the value of slowness while also confronting its costs. The narrative that "it worked out in the end" risks dismissing those who died in the interim.
The EU AI Act: 3 years. The thirty-six-hour marathon negotiation we saw in Chapter 4. From the initial proposal in 2021 to final approval in 2024 — three years. Fast by historical standards. Slow by the standards of technological change — during the same period, ChatGPT launched, generative AI reshaped the world, and China implemented its generative AI regulations in roughly eight and a half months (ChatGPT launched November 30, 2022; China's Interim Measures for the Management of Generative AI Services took effect August 15, 2023).
The mechanism by which slow justice gains resilience: evidence-based decision-making (the Sadler Committee's 682 pages, EU impact assessments). Iterative learning (from the failure of the 1802 Act to the success of the 1833 Act). Pluralistic checks (tribunes, bicameral legislatures, the judiciary). Incremental expansion (from cotton mills to all manufacturing).
The result: short-term pain, long-term stability.
Regime longevity proves the point. British parliamentary democracy: 338 years (1688–present). The American constitutional order: 249 years (1776–present). Napoleon's regime: 11 years. The Chinese Communist Party: 77 years (ongoing). The regimes of fast order are short-lived; the regimes of slow justice endure.
3. Five Patterns — and One New Question
This book has traced a single question across twelve chapters. The gap between the speed at which technology transforms society and the speed at which institutions respond to that transformation — what is the structure of that gap?
We discovered five patterns. But by Chapter 12, the question to ask is not whether the patterns exist. It is whether they are accelerating. The same structures recur, but the timescale on which they operate is compressing.
First, information asymmetry — its depth has changed. The structure in which the reality of harm never reaches institutions. In Chapter 1, the small farmers could not write and had no seats in the Senate — yet the latifundia were at least visible. Pliny could record the size of the great estates. In Chapter 7, Driver Lee could not know how the dispatch algorithm determined his assignments, routes, and ratings — algorithms are invisible. In Chapter 8, Mr. Park received no explanation for why AI denied his loan. Transparent data, opaque judgment. The MyData system (마이데이터) integrated financial information seamlessly, but no one bore any obligation to explain how the AI interpreted that information to arrive at a denial. Rome's information asymmetry was a problem of access. The information asymmetry of the AI age is a problem of comprehension — when even the designers cannot explain the judgment, how can the harmed party contest it?
The deliberative democracy and digital participation platforms of Chapter 11 were a response to this asymmetry. What Taiwan's vTaiwan demonstrated, what Audrey Tang meant when she said "democracy is a technology" — an attempt to reverse information asymmetry through technology itself.
Second, incumbent capture — its form has grown more sophisticated. The structure in which the regulated become the regulators. In Chapter 1, the three hundred senators simultaneously owned latifundia and held legislative power — direct capture. In Chapter 6, Big Tech companies monopolized technical advisory roles at congressional hearings, creating a structure in which they effectively said, "Please regulate us this way." The licensing regime that Sam Altman requested before the Senate in Chapter 6 was an entry barrier that favored firms already possessing large models — expertise capture. Because legislators did not understand the technology, the regulated became the experts who designed the regulation. In Chapter 10, even RegTech — the attempt to automate regulation through technology — risked capture by the same companies. When you try to regulate with technology, the entity you are trying to regulate turns out to own the technology.
Third, the cost of consensus — the timescale has changed. The paradox in which democratic checks block innovation. In Chapter 1, a single tribune's veto blocked Gracchus's entire agrarian law. In Chapter 2, British factory legislation took sixty-four years. In Chapter 4, the EU AI Act required three years of negotiation capped by a thirty-six-hour marathon. In Chapter 11, deliberative democracy tried to reduce the cost of consensus through digital technology — as when the Shin-Kori nuclear reactor public deliberation (신고리 5·6호기 공론화) reached a conclusion in three months with a citizens' jury of 471 members.
But reducing the cost of consensus is different from bypassing consensus altogether. China implemented its generative AI regulations in roughly nine months not because the cost of consensus was low, but because the process of consensus did not exist.
Fourth, the gradualism of crisis — it has accelerated. The phenomenon in which cumulative catastrophe never exceeds the detection threshold at any given moment. In Chapter 1, the decline of the small farmers unfolded over decades. To each generation, the crisis was a change that was "still bearable." In Chapter 3, the risks within the financial system were "bearable" — until Lehman Brothers went bankrupt in 2008. In Chapter 9, AI-driven job displacement was advancing gradually — 240 call-center agents received layoff notices by text message on a Friday afternoon, 70 percent of newly certified CPAs could not find employment, junior developer hiring was shrinking — but each of these, taken individually, did not yet register in macroeconomic employment indicators. Translators declining, customer service agents declining, data-entry clerks declining — none individually constituted a dramatic event. Yet the decline of Roman smallholders unfolded over 130 years; AI displacement needs only 130 months. The structure of gradualism is the same, but the unit of gradual change has shifted from generations to quarters.
Fifth, ideological barriers — the same sentences repeat. The belief system that blocks the very legitimacy of intervention. In Chapter 1, the custom of the ancestors (mos maiorum) served as a shield to attack Gracchus's reforms as "destruction of tradition." In Chapter 3, Alan Greenspan's conviction that "markets correct themselves" delayed financial regulation for thirty years. In Chapter 6, the ideology of "don't hamper innovation" justified the absence of federal AI regulation in the United States. In Chapter 8, the myth of "technological neutrality" blocked institutional intervention against algorithmic discrimination. The words differ, but the structure is identical — "This is the way it has always been done," "The market regulates itself," "Regulation kills innovation." All that changes from era to era is the vocabulary.
The five patterns reinforce one another. Ideology legitimizes incumbent power; incumbent power raises the cost of consensus; high consensus costs preserve information asymmetry; information asymmetry obscures the gradualism of crisis. Two thousand years ago and today, only the names have changed. The Senate became the board of directors. The latifundium became the platform. The chair leg became the algorithm.
But one thing has changed. Speed. Where the spread of the latifundia took 130 years, the spread of AI has taken fewer than 130 months. The structure of institutional delay is the same, but because the speed of technological change has increased exponentially, the gap is widening to an unprecedented degree.
Earlier in this book, we saw the "illusion of coexistence" accelerating. Roman smallholders believed for a hundred years that their way of life would continue. Handloom weavers during the Industrial Revolution held out for forty years. For those displaced in the AI age, the window was two years. If the period of illusion has shrunk from a hundred years to two, the response time of institutions must shrink accordingly. Institutional adaptation that once took fourteen to sixty-four years must now happen in two — this is why "a better speed" is needed.
4. The Return of the Displaced
Let us return to Mr. Park.
In Chapter 8, he was denied a loan by an AI credit-scoring system. He never learned why. That night, in the back room of his shop, he reopened the rejection screen. "Comprehensive credit assessment: ineligible." He had been running his business for twenty-four years. He had never been late on taxes, never missed a payment to a supplier. Against a judgment that a machine had rendered in three seconds, there was nowhere he could file an objection. In Chapter 10, he caught a faint glimpse of what regulatory technology (RegTech) might make possible — a world in which AI monitors AI.
Now we ask: Is a world in which Mr. Park can challenge an AI credit assessment possible?
History says yes — it just takes time. In 2023, the U.S. Consumer Financial Protection Bureau (CFPB) issued guidance stating that "black-box AI models cannot be used as a means to evade credit denial explanation requirements." The response Mr. Park received in Seoul — "It's a system result, so we don't know the reason" — was beginning to be rendered unlawful in Washington. South Korea's AI Basic Act (AI 기본법, effective January 22, 2026) imposed explanation obligations on AI developers — but not on the banks that deploy AI. The law exists, yet there remain places it does not reach. Just as Gracchus's agrarian law existed but was never enforced.
Let us return to Driver Lee.
In Chapters 7 and 9, we saw his daily reality — a day governed by the dispatch algorithm, a 0.68 percent employment insurance claim rate among platform workers. A new scene is laid atop that reality.
Now we ask: Is a world in which citizens can oversee delivery algorithms possible?
In July 2024, the Supreme Court of Korea (대법원) ruled in the TADA case (타다 판결) to recognize the worker status of platform laborers for the first time. That same month, a popular delivery rider who had been documenting fifteen- to seventeen-hour workdays on YouTube was killed in a traffic accident. The ruling and the death occurred in the same month. How would Driver Lee have heard the news? Probably through a smartphone notification while waiting for his next dispatch. And thirty seconds later, the next dispatch alert would have sounded — the algorithm does not allocate time for mourning. In the time it took institutions to take a single step, reality had already claimed a life. Those who pay the highest price for slow justice are the displaced.
Let us return to CEO Choi.
In Chapter 10, she was running an AI medical imaging startup and had received three different legal opinions from three law firms. "Is this a medical device or software?" — each firm gave a different answer. Legal fees: 12 million won (approximately $9,000). Of her 500-million-won seed investment (approximately $375,000), 2.4 percent had been spent "finding out what the law is." In the financial sector alone, there were some forty AI-related guidelines, and applications for the Innovative Financial Services program (혁신금융서비스, a regulatory sandbox) hit an all-time high of 436 in 2024 — a figure that was itself proof that companies could not be certain whether their businesses were legal.
Now we ask: What does the world look like after succeeding in a regulatory sandbox?
CEO Choi's dilemma is the mirror image of Mr. Park's. Mr. Park suffers in a place where the law exists but does not reach. CEO Choi suffers in a place where she cannot find where the law is. A thirty-two-year-old founder reading the third legal opinion at two in the morning. The first law firm said "medical device." The second said "software." The third said "no precedent exists; we cannot advise." Of her 500-million-won seed funding, she had spent 12 million won "finding out what the law is." "The absence of law isn't freedom. It's lawlessness." Her words are not fictional dialogue but a summary of structural reality.
And the three people we met in Chapter 9.
The call-center agent. Twenty years of experience undone by a single word: "efficiency." She was enlisted to correct AI errors, personally producing the training data — then replaced by the very AI she had trained. "We were the ones who trained the AI" — her words reached no institution anywhere.
The newly certified CPA. After five years of exam preparation, he received his certificate, only to find no firm willing to hire him because senior accountants had calculated that "AI is cheaper than training a first-year staff accountant." "Was I born at the wrong time?" The education system was producing workers at yesterday's pace while the market was shrinking demand at the speed of AI.
The junior developer. Companies choosing a 300,000-won-per-month AI coding assistant (roughly $225) over hiring an entry-level developer at 50 million won per year (roughly $37,500). The fastest displacement cycle yet: someone who had learned technology being replaced by the technology itself.
Who, in the end, weaves the safety net for these six people?
Augustus would have had a simple answer. One person decides, one person executes. Fast. But when that person is gone, the system falters. Napoleon would have drafted a code — organize everything in 2,281 articles. But the regime that wrote the code vanished in eleven years.
What these six people need is not fast order. Nor can they wait 240 years.
5. Neither Fast Order Nor Slow Justice — Five Design Principles
It is time to find an answer to the question posed by this book's title.
Fast order: it has speed but lacks legitimacy. Efficient because it bypasses consensus, yet fragile because it has no capacity for self-correction. Augustus's Empire brought 207 years of peace but closed the Republic forever. Napoleon's Code lives on 220 years later, but the regime that produced it died in eleven. China's Reform and Opening Up multiplied GDP 125-fold but could not erase the memory of Tiananmen.
Slow justice: it has legitimacy but lacks speed. It gathers evidence, deliberates, votes, and iterates. The British Parliament achieved universal suffrage over 240 years, and the system has now endured for 338. The U.S. Constitution has been amended only twenty-seven times in 247 years, and the constitutional order has lasted 249. But during those 240 years, during those 89 years, during those 64 years — real human beings lived in suffering.
In the age of AI, neither is sufficient.
In Chapter 10, we saw regulation through technology — adaptive regulation, RegTech, regulatory sandboxes. Attempts to increase speed. In Chapter 11, we saw citizen participation — deliberative democracy, digital participation platforms, vTaiwan. Attempts to secure legitimacy.
When these two are combined, a third possibility opens.
Let us call it "a better speed." A better speed is not an abstract declaration. It is a concrete institutional design addressing each of the five delay mechanisms.
First, a response to information asymmetry — mandatory algorithmic impact assessments. Not the Sadler Committee's 682 pages, but evidence collection based on real-time data. Require independent impact assessments before the deployment of high-risk AI systems, and make the results public. The conformity assessments that the EU AI Act requires for high-risk AI are a starting point. A world in which the reason for Mr. Park's loan denial is not "a trade secret" but "a matter of public record." Just as the Sadler Committee opened the doors of factories, algorithmic audits open the doors of black boxes.
Second, a response to incumbent capture — an independent AI regulatory body. Just as Britain's four factory inspectors of 1833 marked a turning point in the history of regulation, the AI age requires a specialized regulatory body separate from those it regulates. Just as the U.S. CFPB was designed to be independent from Wall Street, an AI regulatory body must be independent from Big Tech's expertise capture. Not the structure of Chapter 6, in which Altman asked "Please regulate us," but a structure in which "We regulate you."
Third, a response to the cost of consensus — institutionalizing digital deliberation platforms. Do not bypass consensus; reduce its cost through digital technology. What vTaiwan's Pol.is demonstrated — not a battlefield of for-and-against, but a tool that first maps the terrain of agreement. The Shin-Kori public deliberation succeeded because there was a prior commitment to accept the outcome. AI policy likewise requires an institutional link that gives binding force to the results of citizen deliberation. To avoid repeating Iceland's failure — where Parliament rejected a crowdsourced constitution that 67 percent of voters had approved.
Fourth, a response to the gradualism of crisis — real-time monitoring of AI's impact on employment. Compressing Britain's "law → failure → better law" cycle to monthly intervals through iterative learning. A leading-indicator system that detects the layoff of 240 call-center agents before it appears in macroeconomic employment statistics. In its 2023 Employment Outlook report, the OECD noted that 27 percent of jobs across member countries fell into the high-risk category for automation and recommended transition policies based on leading indicators.11 The core of adaptive regulation (Chapter 10) is not "regulate and neglect" but the cycle of "regulate, monitor, revise." If the regulatory sandbox is the space for experimentation, real-time monitoring is the experiment's safety mechanism.
Fifth, a response to ideological barriers — reframing "innovation vs. regulation." Not "regulation kills innovation," but "regulation makes innovation sustainable." In Chapter 2, British factory legislation banned child labor without killing the Industrial Revolution — healthier workers in fact produced higher productivity. Just as the ideology of "freedom of contract" gave way to the frame of "child protection," the ideology of "protecting innovation" must give way to "innovation that includes the protection of citizens."
The speed China demonstrated in its 2021 tech crackdown — a $2.8 billion fine on Alibaba, the Didi app deleted within a day, the dismantling of predatory peer-to-peer lending structures — was an empirical case of consumer protection moving faster than democracy allows. Yet the same system enabled the surveillance of ethnic minorities and political persecution. A structure that produces legitimate outcomes through illegitimate means. This is the fundamental limitation of fast order.
Democratic systems, too, can move fast. The COVID-19 pandemic proved it — in the United States alone, 85 emergency rules were published in the Federal Register without public comment during the first four months of the pandemic.12 The Sarbanes-Oxley Act was signed eight months after Enron's bankruptcy (December 2, 2001, to July 30, 2002).13 The Montreal Protocol took thirteen years from the 1974 Molina-Rowland paper identifying the mechanism of ozone depletion to the signatures of forty-six nations (September 16, 1987); it has since been ratified by 198 parties (197 countries and the EU), making it the first treaty in UN history to achieve universal ratification.14 And the agreement worked — the ozone layer is actually recovering, and it is regarded as "the most successful environmental treaty in history."
The conditions for fast justice can be derived inductively. Visibility of harm — the ozone hole was something anyone could see; Enron's bankruptcy was a front-page headline. The existence of alternative technologies or solutions — the Montreal Protocol was possible because CFC substitutes existed. A small number of clearly identifiable actors — the more defined the regulatory target, the lower the cost of consensus. And political will.
The crisis of the AI age is decisively lacking one of these conditions. Visibility of harm. AI-driven job displacement does not appear in the sky like the ozone hole. Two hundred and forty call-center agents, unemployable CPA graduates, declining junior developer hiring — each occurs below the decimal point of the statistics. The gradualism of crisis obscures the visibility of harm. Breaking this structure — making harm visible and designing institutions that respond to visible harm in real time — is the core of a better speed.
6. A Better Speed
Justice is slow. That is why it is precious.
The fastest order was born at the tip of a dictator's pen. The day Augustus "returned power" to the Senate, the night Napoleon overrode the jurists at the Conseil d'État, the December when Deng Xiaoping declared he would "cross the river by feeling the stones." Their speed was extraordinary, and what they produced changed history. But the systems they built — one-man rule beneath the shell of a republic, eleven years of life beneath a brilliant legal code, the memory of Tiananmen beneath an economic miracle — repeated the same structural vulnerability.
The slowest justice grew from the voices of citizens. The British Parliament extended the franchise over 240 years. The U.S. Constitution was amended only twenty-seven times in 247 years. Factory legislation took sixty-four years. And within that slowness, tens of thousands of children worked sixteen-hour days, millions lived in slavery, and 880,000 platform workers remain in the blind spot of a 0.68 percent employment insurance claim rate.
What we need is not faster order. It is a better speed.
What is a better speed? For each of the six people, it is a concrete change tomorrow morning.
A speed at which Mr. Park can check, in real time, the reasons for his AI credit-assessment denial — tomorrow morning, a notification appears on his smartphone: "Credit Assessment Results: 3 reasons for denial. File an objection here."
A speed at which a citizens' oversight panel reviews the criteria by which Driver Lee's dispatch algorithm determines routes every quarter — tomorrow morning, he sees on the app: "Today's dispatch criteria: distance 40%, rating 30%, wait time 30%," and choosing to slow down for safety in the rain does not cost him his rating.
A speed at which CEO Choi can innovate within a regulatory sandbox while patient safety is monitored in real time — tomorrow morning, she checks a single integrated regulatory portal: "AI medical imaging: classified as Software as a Medical Device (SaMD). Applicable standards upon sandbox approval." No need to commission opinions from three law firms.
A speed at which retraining programs and safety nets activate before the call-center agent's layoff is finalized — tomorrow morning, an AI employment-impact monitoring system detects a sharp rise in automation rates in her sector, and transition training program notices are delivered to both the company and the workers.
A speed at which, instead of telling the CPA graduate "you were born at the wrong time," the education system adapts to market changes in three-month cycles rather than three-year cycles — tomorrow morning, the accounting curriculum adds "AI audit tool applications" and "data analytics–based advisory," so that the certificate becomes not a relic of the past but the starting point of a new role.
A speed at which the junior developer, rather than being replaced by an AI coding tool, is trained for a new role collaborating with AI — tomorrow morning, job postings for "AI-augmented software engineer" replace postings for "junior developer." Not displacement but transition.
This is not utopia. As the adaptive regulation of Chapter 10 and the deliberative democracy of Chapter 11 have shown, the technological means already exist. The Montreal Protocol proved it — when harm is visible, alternatives exist, and actors are clearly identifiable, democracy can move fast. What is lacking is not technology but the choice. And the choice is still open.
In Book 2, we met Sarah in America and Wang Lei in China. One falls freely, the other stagnates under control. And in both countries, the institutional redesign that would serve them has not yet begun. The word "yet" matters — because it is not a declaration of failure but a declaration of possibility.
We return to the opening of Chapter 1.
The summer of 133 BC. The Capitoline Hill. Senators charging with chair legs raised. Weapons made not of marble but of wood. The sound of institutions destroying institutions.
In Chapter 6, we saw the modern version of that hill. A hearing room in Washington, with mahogany panels and white marble columns. Altman said, "Please regulate us." Lawmakers asked, "What is ChatGPT?" Instead of chair legs, lobbying dollars were the weapons. No blood was shed, but the structure was the same — in a room where regulator and regulated had switched places, slow justice was once again being defeated. But on the Capitoline, people died. In Washington, no one has died yet. Yet.
What Gracchus wanted was simple. To enforce a law that had gone unenforced for 234 years. It was an attempt at slow justice. When it was crushed, Rome accepted Augustus's fast order after a hundred years of civil war.
Two thousand one hundred and fifty-nine years later, we stand at the same fork in the road.
A world in which AI denies a loan in three seconds, recalculates a delivery route every 0.3 seconds, and sends layoff notices to 240 smartphones on a Friday afternoon. Technology moves faster than human institutions can follow. Faced with this gap, someone will propose fast order — an order that bypasses consensus, suppresses dissent, and concentrates decision-making power in the hands of the few in the name of efficiency.
Gracchus had to run. We do not have to run. Not yet.
But to avoid running, we must walk at a better speed. A speed that does not bypass consensus but lowers its cost. A speed that gathers evidence — not across 682 pages but in real time. A speed that, when a law fails, produces a better law in three months rather than 240 years. A speed at which the voices of the displaced reach institutions.
Justice is slow because justice listens. Order is fast because order does not listen.
We can listen — and listen faster.
At four in the morning, Mr. Park turns off the light in the back room of his shop and rises. Tomorrow morning, he will open the banking app again. This time, the reasons for the denial might be written there. A "file an objection" button might have appeared. The things he has built over twenty-four years — his tax payment record, his supplier payment history, the seasonal patterns of his revenue — might be restored as context the machine had been unable to read. Not yet. But "not yet" means there is still a "yet."
At the same hour before dawn, Driver Lee starts his motorcycle. The first dispatch alert sounds. It is raining again today. But someday — the day will come when the algorithm recognizes "safe deceleration on a rainy day" not as a penalty but as the norm. That day will be the day institutions began to listen.
When institutions begin to listen, slow justice is no longer slow.
This is the conclusion of this book. And this conclusion is not an ending. In the next volume, we ask — in a world where the speed of institutions determines the fate of nations, what speed will Korea choose?