Democracies Need Creative Destruction

Over the past decade, there has been a growing consensus that democracies around the world are facing a crisis from within. According to a 2025 Pew survey, nearly 60 percent of respondents in 23 countries agreed that they are “dissatisfied with democracy.” Polarization, disinformation, and disillusionment with leaders willing to break established norms for personal advantage are certainly to blame for this growing sense of dysfunction, but there is a deeper flaw in today’s democratic governments: state agencies and institutions are increasingly unable to perform the basic tasks they were designed to carry out.

When the COVID-19 pandemic reached the United States, in early 2020, the U.S. government was spending roughly $600 million a year to detect infectious diseases. Yet the virus spread undetected for a month because researchers were unable to get permission from a rigid bureaucracy to test the swabs that would have helped identify it. In the United Kingdom, despite the billions of dollars the government has spent on youth employment services following the pandemic, the number of young people aged 16 to 24 who are not employed, in school, or in training programs has surpassed one million—a precipitous increase of more than 300,000 in less than five years. Germany, a country once known for its efficiency, now has a rail network so unreliable that it has become a national embarrassment; around 40 percent of long-distance trains arrive late. Across large swaths of the United States and Europe, people compare what they hear from politicians of all parties with what they see with their own eyes, and the rhetoric does not match the reality. The result is dismay and distrust.

“State capacity” used to be a term that experts in advanced countries used when speaking of the developing world. If this country receives foreign aid, does it have the capacity to turn it into public good? Now the citizens of those donor countries are asking that question at home. If we pay taxes, will our government provide more of the things we need? If we engage in the political process, will we get the outcomes we seek? Increasingly, the apparent answer to these questions is no. As Brian Shearer, a regulatory expert at the Vanderbilt Policy Accelerator, has laid out, implementing a significant new federal rule in the United States usually takes at least five years from start to finish, making it impossible for a president to deliver on a promised reform before the next election. As a result, the basic feedback loop of democracy—candidates say what they’ll do, voters choose, results arrive, voters judge—is breaking down. Nearly all rich democracies face some version of this crisis.

At a fundamental level, these failures not only erode confidence in democracy; they also undercut sovereignty. When Ukraine needed to restore its communications network after Russia’s invasion, in February 2022, it turned to Starlink’s satellite Internet service, a private enterprise. That meant that although the Pentagon eventually paid to keep the Starlink network running in Ukraine, the decision about whether Ukrainian forces would have the satellite network they needed to strike the Russian fleet at Sevastopol rested not with the Pentagon or with the National Security Council but with Starlink’s founder, the tech billionaire Elon Musk. Since then, frontier AI models have only accelerated the shift of power from public institutions to private companies.

Democracies have not given up on institutional reform. There are glimmers of success in almost every advanced country, including the United States, at agencies such as the Internal Revenue Service, which has finally begun to modernize its woefully out-of-date technology. Following the onset of the pandemic, the massive stimulus that Congress passed in March 2020 showed what governments could deliver when the global economy was on the line. But overall, efforts to make government agencies fit for purpose in today’s world have not been going well. Even Musk, despite his vow to take a chainsaw to the bureaucracy, seemed exhausted by the vastness of what his Department of Government Efficiency had taken on, and abandoned DOGE over a year earlier than planned.

To truly reverse the dramatic decline in state capacity, governments need to take radical approaches that can win public trust. In his classic 1967 book, Inside Bureaucracy, the economist Anthony Downs described how institutions that lose their ability to adapt eventually embark on a slow death, becoming hollow shells of rules and procedures. Once they reach this point, he argued, they must be either reborn or destroyed. Today, advanced democracies around the world need to give serious attention to institutional rebirth—rebuilding state agencies into departments of now, unencumbered by people, precepts, and processes of past eras. But they also need to consider which legacy agencies are beyond reinvention and should be shut down altogether.

Not every broken institution has to be rebuilt from scratch. Governments do not have the time, resources, or political bandwidth to turn over the tables in every part of the state at once, nor would doing so be wise. Relentless, incremental improvement of existing agencies remains an important part of statecraft. But if they are going to solve the current crisis, democracies will have to get comfortable with creating new institutions. Curing what ails democracy today will require many things, but no other remedies will stick if governments can’t deliver what they promise.

TEST, LEARN, REPEAT

There is no single correct formula for institutional rebirth. An institution might be considered reborn when it has installed a new operating model that allows it to achieve its stated mission. Some legacy institutions hold valuable assets—authorities, data, know-how—that are not easy to move to a new home, so governments need to figure out how much of the existing system is irretrievably broken and what if anything needs to be preserved. It is critical to get the degree of newness right.

In North America and Western Europe, state institutions have inherited an operating model that was built in the era leading up to and immediately following World War II. This meant that they were designed to administer large, complicated, stable programs through careful planning and centralized control and that the problems they were supposed to address were assumed to be known in advance. During the postwar decades, that model was generally workable, but when the Internet began to rewire the economy in the late 1990s, it introduced a very different set of practices. Solving problems now depended on a system of testing and learning in which solutions were continually tried out in the real world and adapted accordingly. Today, with the advent of AI, the pace of scientific and technological change has only increased.

As a result, legacy state institutions must deal with citizens who expect services to be tailored to their needs, and they must do so in a world that has itself dramatically sped up. And these institutions are constrained by decades of procedural accumulation—rules and requirements added in response to yesterday’s problems and rarely removed even after the problem or the technology has changed. It is a dynamic that increasingly leads to paralysis. Only by adopting new operating models can they begin to address the challenge. Sometimes, this can be achieved without building a completely new entity; in other cases, a clean break is required. Either way, the people running the new operation must have the freedom to define how they work, whom they hire, how they buy, and how they account for themselves to the public. Otherwise, they will just end up replicating what they are trying to replace.

The clearest case for starting from scratch is when the goal or objective is also new. In 1958, for example, the U.S. Congress created NASA because beating the Soviet Union to the moon was a task that could not plausibly be handed to any existing military or civilian agency. But building a brand new entity may also be warranted when an existing institution can no longer adequately perform a well-understood task. In 1997, South Africa replaced two underperforming tax authorities with a wholly new revenue collection agency. By 2022, the new agency had brought in tens of billions of dollars with revenues reaching an annual compound growth rate of 9.9 percent, allowing the post-apartheid state to fund a significant expansion of services. The Australian state of New South Wales did much the same in the 2010s, when it brought the poorly organized functions of 16 public-facing agencies into a single entity with a unified digital platform. The new agency rapidly transformed public attitudes about government services; the number of people who reported they were satisfied rose from less than 60 percent to over 97 percent during its first five years of operation.

Another effective approach is to give a new entity a specific task that falls under the purview of an existing agency that is unable to perform it. In 1958, after the Soviets’ successful launch of Sputnik exposed how interservice rivalry in the U.S. military had left missile and rocket programs duplicative, slow, and uncoordinated, the U.S. government pulled advanced research out of the individual branches and created the Advanced Research Projects Agency—ARPA, later renamed DARPA—to run it independently. Other defense research functions that sat well within the existing hierarchy, such as mission-specific R & D tied to each service’s existing weapons systems, stayed within the branches. The DARPA approach worked so well that it spawned an entire family of similar entities, such as the Advanced Research Projects Agency-Energy (ARPA-E), to do research for the Department of Energy, and the United Kingdom’s Advanced Research and Invention Agency (ARIA).

Sometimes, the place to build a new institution is inside the one that is broken. In 2019, the state of Colorado put small teams of engineers, designers, and product managers—part of a new entity called the Colorado Digital Service—in place alongside existing staff in state agencies. Since then, the CDS teams have introduced each agency to a test-and-learn model for developing the software it needs, and the changes have slowly built demand for the new practices across the entire state bureaucracy. In May 2026, Colorado announced that its Office of Information Technology, a legacy institution with a staff of 1,150 people, would switch to the CDS’s operating model and that the head of the CDS would take over as the new director. It might seem surprising that what amounts to a “skunkworks” could take over a critical IT function of a midsize U.S. state that has an annual budget of $46 billion, but the heads of other Colorado agencies welcomed the shift, recognizing that the legacy model was no longer able to deliver.

Yet another approach emerged in the 2013 reset of Universal Credit, the United Kingdom’s new welfare payment system. In this case, the rebirth strategy involved no new organization or institution at all. Combining six existing working-age benefits into a single payment, Universal Credit had been adopted as a major welfare reform three years earlier, only to produce what a Parliamentary committee described as “extraordinarily poor” results. So the minister responsible, Iain Duncan Smith, installed new leadership inside the part of the Department for Work and Pensions that was overseeing Universal Credit and gave it a new mandate. Adopting a test-and-learn approach, the new leadership team began small, but over time, every aspect of the agency’s work—its policy, its operations, its technology—was tweaked and adapted. Eventually, the new approach enabled Universal Credit to be rolled out to more than eight million claimants and to keep functioning even amid the surge in demand during the COVID-19 pandemic. The program also showed marked improvement over the existing system in helping single parents get back into the workforce, according to government figures. This was all done within the same department, with the same statutory purpose.

When an existing institution holds genuine assets that cannot be rebuilt from scratch, working from within can be the best option. Some know-how is too precious or too dangerous to risk rupturing: think of the people who look after the U.S. nuclear arsenal. Moreover, although data and legal authorities can be transferred between bodies, doing so may be prohibitively time-consuming. But it is important to recognize that transforming an agency from within is not necessarily any less difficult, given the necessary proximity of the new team to the legacy ways of working. Some of the assumed advantages to this approach—that a successful pilot project within the agency will naturally spread the new operating models to adjacent teams and functions, for instance—often don’t hold up in practice. Consider what happened at the Internal Revenue Service in 2024. Using a test-and-learn model, a skunkworks team built Direct File, a system that allowed U.S. taxpayers to file their returns electronically, in less than a year. Rather than embrace the new practices that the team had introduced and leadership had supported, the agency largely treated them as a threat and, if anything, dug in to the status quo.

HOUSTON, WE HAVE A SOLUTION

According to a well-worn anecdote, U.S. President John F. Kennedy once asked a janitor working for NASA what his job was. “Mr. President,” he responded, “I’m helping put a man on the moon.” Apocryphal or not, the story captures the most critical element of any successful new institution in government: every single person working there must be exceptionally clear about the mission.

As daunting as it may sound, it is quite possible for an agency or institution to set and achieve clear, meaningful, and ambitious goals. When Universal Credit was reset in 2013, after three years of failure, the new team asked Smith, the elected official accountable for delivery, “What is the outcome you actually want to achieve?” His answer was not “deliver Universal Credit.” His answer was “support more people, to find more work, more of the time, while protecting those who can’t.” The team wrote that down and stuck it on the wall; it became the North Star for the turnaround that followed.

At the same time, new and highly disruptive institutions must also have extremely visible political backing. During the pandemic, British Prime Minister Boris Johnson’s full-throated support for the government’s Vaccine Taskforce and his unambiguous directive to “stop people from dying” were essential ingredients in the group’s success. In other cases, some of the most effective support for new institutions comes from officials who bring other forms of authority. As the sponsoring minister for the new Government Digital Service in the United Kingdom between 2011 and 2015, Lord Francis Maude was not the most senior person at the cabinet table. But his personal authority, reputation, and long tenure made him a highly effective advocate for the agency, helping it transform several large public services. Partly as a result, the United Nations ranked the United Kingdom as the number one e-government in 2016.

The U.S. Capitol, Washington, D.C., July 2026 Al Drago / Reuters

Ultimately, however, success will depend on the new entity’s ability to do its assigned task. The performance of the founding team will likely determine whether the institution manages to escape bureaucratic traps or reconstitute them under a new name. To succeed, the team will need policy experts who understand legislative intent, operational staff who know how government truly functions, and technologists—a multidisciplinary mix of personnel who all take a test-and-learn approach. Ideally, the team should include insiders as well as outsiders: too many career officials will stifle disruption, but too many newcomers who are ignorant of how bureaucracy works can also be fatal. Especially useful will be a mix of hybrid characters: insiders who chafe against institutional norms paired with outsiders who are drawn by public purpose.

It’s easy to miss how important it is to get the support functions right. At DARPA, the reverence for program managers is readily apparent. The agency prides itself on the talent it attracts in these roles, and new institutions often strive for that same focus on specialized knowledge. But it is the basics that doom most efforts to do things differently. If it takes six months to hire someone or three months to get memos to the senior executives, whatever exceptional talent the agency can attract will be wasted. At DARPA, program managers hold their positions for four years, but in some government agencies, it’s common for a single procurement to take that long. DARPA’s procurement function is forced to keep up with the ambitions of its talented experts.

To establish the right operational DNA, institution builders should select people with a history of running toward what is new and difficult rather than those who have already proved themselves in legacy organizations. Backroom staff who support hiring, procurement, finding computers and office space, contending with internal governance, and so on, need to have the same innovative, bootstrapping mindset as strategists and policy experts, or the institution may find itself hamstrung by operational constraints regardless of its other strengths. Choosing the right leadership is a similar challenge. The most successful leaders more closely resemble startup founders than agency heads: in other words, people who are comfortable with building from scratch. There are exceptions. Sometimes, institutional veterans are radicalized by their frustration with the status quo. But a vendetta alone won’t suffice; leaders need a positive vision.

THE POWER OF NOVELTY

The word “institution” conjures an image of permanence. But today’s government agencies must act with great speed and respond to novel threats and challenges. In May 2024, the United Kingdom established the Infected Blood Compensation Authority to provide financial payments to thousands of people who received contaminated blood and blood products from the National Health Service from the 1970s to the early 1990s. At the time the authority was created, an average of two victims a week were dying before they could receive the compensation they were entitled to. Still, the pressure to get payments out had to be balanced with the need to handle complex cases carefully. So IBCA chose a test-and-learn approach, prioritizing certain cohorts of potential recipients and incorporating the lessons of those early interactions into the service as it scaled up. In an assessment of the authority’s first 18 months, an expert review panel reported that more than 2,000 people had received payments totaling 1.36 billion pounds (about $1.83 billion). Had the new team followed the conventional institutional approach, the panel noted, in all likelihood no claims would have yet been processed.

The IBCA’s impressive efficiency was possible only because it owned the process from beginning to end. Governments have a strong tendency to break up the delivery of services into an assembly line, starting with policy and legislation and finishing with implementation. New institutions must not get trapped into owning only a single step of a larger sequence. A team that is focused only on writing better policy without the ability to reshape the operating model—or a team that is designed to improve delivery outcomes but lacks the mandate and access to influence the policies that govern that delivery—will ultimately have limited impact.

But creating new institutions is only half the battle. Just as important is determining what to do with the old ones they are designed to replace. Of course, if a legacy institution can be successfully remade, as in the Colorado case, there is nothing to close. But more often, the new institution ends up competing with its still extant predecessor. For bureaucracies to function, governments have to break the self-reinforcing cycle of legacy accumulation, which means turning off old institutions as deliberately as they launch new ones. The U.S. government navigated this challenge well when it created the Consumer Financial Protection Bureau in 2011. Before its founding, seven agencies nominally held responsibility for various elements of consumer protection, but none had it as a primary mission. By consolidating these responsibilities into a single body with a clear mandate and transferring the relevant authorities to the new agency, the CFPB was able to return more than $21 billion to consumers in monetary compensation, canceled debts, and other forms of relief from fraud, before being significantly downsized by the Trump administration in 2025.

By contrast, the Department of Homeland Security shows what can happen when a new body is created without addressing conflicting authorities. When it was set up, in 2003, the DHS consolidated the functions of 22 existing agencies into one. But it never resolved its overlapping counterterrorism role with the FBI and the CIA, and it created new internal overlaps among the agencies it oversees, such as between Immigration and Customs Enforcement and Customs and Border Protection.

Of course, no institution is new for long. As a novel institution starts to make a difference, it has to interact with more of the larger bureaucratic machine, and those connections can sap its disruptive force. Smart design, strong leadership, and political support can ward off this effect, but diminishing returns over time is inevitable. Even so, new institutions are still worth the effort. The inevitable entropy of institutions is itself a strong reason to become better at creating new ones; expecting an indefinite lifetime of high performance is apt to disappoint. And new institutions don’t have to last to be worth building—a burst of capacity, however temporary, can produce results that outlast the institution itself. The creation of NASA arguably amounted to only a temporary boost to U.S. state capacity. But without that lift, the country would not have put a man on the moon.

That is the tactical argument for creating new institutions. The strategic argument is that by showing what a temporary increase in state capacity can do, a successful new entity can make the case for the deeper structural changes that governments need to achieve a permanent increase. New institutions get their room to maneuver through carve-outs from rules that are meant to apply across all of government, such as hiring authorities that bypass the normal process or waivers from cumbersome procedural requirements. But the civil service rules, procurement and procedural statutes, and oversight frameworks those carve-outs are designed to skirt are themselves changeable. If governments were better at reforming them properly and more often, they could enable many institutions to meet the public’s needs without requiring so much rebirth.

Such structural change is possible. Most developed countries have already done it once, when merit-based civil service reforms dismantled the patronage-driven spoils system over a century ago. But the political case for it can be very hard to make. As long as the benefits remain theoretical, the exercise of updating the operating model of government may seem dry and technocratic. A new institution that is visibly delivering on its purpose, however, turns an abstract argument about incentives and rules into a concrete question: Why not change those deep defaults—how public servants are rewarded, how accountability flows around the political system, how the state and citizens interact—throughout the government?

A FRONTIER MODEL FOR DEMOCRACY

As many countries wrestle with declining public confidence in democracy, political attention is returning to the craft of governing. But to arrest that decline, governments will have to address the faltering state capacity that is driving it. And that means being prepared to jettison legacy institutions that are beyond repair. It is difficult to imagine, even with the most brilliant officials in charge, a high-performing Federal Emergency Management Agency in the United States or Home Office in the United Kingdom. The same goes for thousands of other ministries and public bodies in these and other advanced democracies. Many have been dubbed “not fit for purpose” repeatedly over several decades.

But governments will also have to proceed with care: the cost of getting this wrong is rising. Along with DOGE in the United States, there is the example of the United Kingdom’s Mission Delivery Unit, also created in 2025, and charged with driving Prime Minister Keir Starmer’s five top political priorities through the system and changing the existing machinery of state. Yet neither delivered what was hoped for, whether in accomplishing the goals it set for itself or in rebuilding public confidence in institutions. If loud promises to change the state become associated in the public mind with willful destruction and negligible results, better-planned efforts will have to contend with even greater public cynicism and impatience.

Moreover, time is short. AI technologies are already probing the seams of existing state bureaucracies and will not wait for governments to find their footing. Chris Schmitz, a researcher at the Hertie School’s Center for Digital Governance, has coined the term “agentic flooding” to explain the mounting demand on public services driven by citizens using AI agents—tools that draft appeals, fill out complex applications, and navigate bureaucratic processes on their behalf. But overloaded services are not the only thing at stake. When agencies lack the technical staff to write or enforce AI-specific rules—on model safety, algorithmic hiring, content moderation, autonomous systems—the practical rules get set by privately owned AI labs and platforms, speeding up the transfer of sovereignty to nonstate companies and foreign actors.

New institutions, even those that are built and sequenced properly, will not be a panacea. But they are one mechanism that democracies can use to plausibly move at the pace this moment requires. A state that tries to perform its usual functions in the AI era but without AI-era institutions is one that will be hastening its own administrative decline. So far, the experience of organizations both public and private suggests that AI is an amplifier. If an organization is built with the right new conditions in place, AI may unlock great potential, freeing up time for frontline staff to deliver public goods and services in new ways. If the organization is a mess, AI will simply make it more dysfunctional, faster. Fourteen years after the Declaration of Independence, the British political philosopher Edmund Burke wrote that “a state without the means of some change is without the means of its conservation.” That is truer now than ever.

Loading...