We Gave Bureaucracy the Fastest Tools in History. It Got Slower.

Every country runs on two clocks. The first is the speed of need: how fast a problem arrives. The second is the speed of red tape: how fast the institutions we built can answer it. For most of the last century the two clocks ran close enough that the gap was an annoyance. It is not an annoyance anymore. It is the gap our rivals are aiming at, and it is the gap most of us feel every week without a name for it.

Start with a department built for speed. In the winter of 2003 the United States created the Department of Homeland Security to answer a new kind of threat faster than the old arrangements could. It opened with three layers of leadership and three people in them: a secretary, a deputy, and an under secretary. A political scientist at Brookings who had spent decades counting senior executive titles across the federal government went back fifteen months later. The new department had twenty-one layers and one hundred and forty-six occupants. Nobody had voted on the resulting architecture as a whole, and no single person designed it. Congress added some, political control added some, growth and promotion added the rest, and each one arrived with a reason. By spring 2004 the department created to move fast had a hierarchy taller than most of the departments it was meant to outrun. The same count found that across all the cabinet departments, the number of distinct senior executive titles had grown from seventeen in 1960 to sixty-four. That is the reflex this article is about. It is not confined to one department, one party, or one era. It is what our systems do when nobody is counting the layers.

We like to talk about national power as four engines: diplomacy, information, the military, and the economy. Strategists call it DIME. What we talk about less is that all four engines draw on the same administrative capacity at home, and that capacity is caked in plaque. Much of it still runs on systems built before most of its workforce was born: in 2019 the Government Accountability Office identified the ten most critical aging systems in the federal government, between eight and fifty-one years old, several written in COBOL, costing a third of a billion dollars a year just to keep running, and six years later two of the eleven systems on the updated list still had no modernization plan. The bottlenecks differ from engine to engine. What they share is that none of them is measured on the same clock, and that the clock is slow.

Take the letters one at a time. Diplomacy: the machinery that seats an ambassador runs through nomination, vetting, committee, and floor, and it has run slowly under both parties. Two years into the previous administration, the diplomats’ own association counted more than forty ambassadorships unfilled, some empty for as long as five years; its tracker has recently hovered around one hundred. Whatever the politics of any given year, the process itself has not kept pace with the world it is supposed to speak to. Information: the contest to understand and shape what people believe moves at the speed of a news cycle, and the machinery for it moves at the speed of a strategy document. The Defense Department published a strategy for information operations in 2016. Three years later the Government Accountability Office found no implementation plan and no investment framework behind it; the department disagreed with the auditors’ recommendations, and Congress had to order a posture review by law. In 2023 the department issued a new strategy, which opens by conceding that America’s competitors are moving quickly in the information environment. Seven years, two strategies, one act of Congress, and the contest did not wait. Military: a major weapon now takes more than twelve years from formal start to a first usable version, for reasons the auditors name precisely, immature technologies at the start, optimistic schedules, rigid requirements, and a sequential process that does not learn as it goes. And its auditors have issued a disclaimer for eight consecutive years because the department could not provide sufficient reliable evidence supporting its agency-wide financial statements. Economic: the instruments of economic power abroad, sanctions, trade, assistance, and defense trade, are administered by the same machinery. Take the sale of arms to allies, a hinge between the economic and military instruments. When an ally buys American weapons, the time to a signed contract is measured in years, not months; the auditors found the department not meeting its own timeliness goals, unable to say how long it takes to deliver the first item, and lacking any central system to track the approvals that gate a sale. The consequence the auditors draw is plain: partners wait for capabilities they need to meet the security objectives we share with them. Four engines, four different bottlenecks, one administrative capacity beneath them, and no shared clock.

Now the part that should unsettle anyone who assumed technology would fix this on its own. Over the same decades, the tools we run the machinery with got faster. Not a little faster. The computer age arrived, then the internet, then cloud software sold on the promise of efficiency, and now artificial intelligence. If the machinery were slow because it lacked processing power, those tools would have closed the gap. They did not.

Let me be exact about what the record shows, because the headline is a provocation and the argument has to earn it. The record does not prove that faster tools made government slower. It shows that the tools got faster, that the institutions did not get faster, and that by the few measures we keep, several got slower. Between 1960 and 2004, while computing moved from punch cards to the web, the count of senior executive titles nearly quadrupled. Between 2012 and 2024, the years in which nearly every office in America moved onto cloud software, the share of American wages devoted to compliance tasks did not fall; it edged up, from 4.0 percent to 4.2 percent. The tools for designing a weapon are incomparably faster than they were in 1990, and the time to deliver one has lengthened in each recent assessment. And one early, visible byproduct of the newest tool, generative AI, is a database of court decisions in which a court found, or implied, that someone had relied on citations the machine invented. The database does not establish AI use or identify the tool in every entry, but it now catalogs more than two thousand decisions in which courts confronted apparently hallucinated material.

I do not know whether the same is true of our rivals. Nobody outside their systems can count their days either, and closed governments do not publish their audit failures. Autocracies can decide fast at the top and badly everywhere else. What I know is that the only clock we control is ours, that the tools arriving now are the fastest in history, and that a system whose speed does not improve as its tools improve is telling you something about where its bottleneck is.

Here is the mechanism, stated as carefully as I can. Tools speed up whatever they touch. They touch drafting, searching, routing, and reporting. In a large institution they mostly do not touch the things that gate a decision: who holds the authority, how many reviews sit between a request and an answer, and who will be blamed if the answer is wrong. When production speeds up and decision rights stay the same, the institution can produce work faster than it can decide. More drafts, more reports, more requests, and more data flow into the same approval queues, and the queues get longer. A faster fax machine makes more faxes. Worse, poorly integrated tools can become one more system that must be fed, reconciled, secured, and audited, which is to say they add a layer of their own. This is not a new observation. In 1987 the economist Robert Solow wrote that you could see the computer age everywhere but in the productivity statistics. What is new is the scale. We are about to hand the most powerful drafting tool ever built to institutions whose bottleneck was never drafting. Unless the decision rights move, the queue can grow, not because the tool failed but because the bottleneck moved.

None of this is waste in the way people usually mean, and not every review is plaque. Some reviews prevent corruption, protect rights, improve safety, or preserve the legitimacy that lets a free government act at all. The question is never whether a process creates delay. It is whether the delay buys protection proportionate to the risk, and whether anyone still knows what the protection is. The cost of the layers that fail that test is not measured in dollars. It is measured in time, and time is the one thing an adversary cannot lend back to you. A weapon delivered in twelve years answers the threat of the year it was conceived. An ally who waits years for a contract remains without a capability it needs. A chair that sits empty for a season is empty through the season when the chair mattered.

You do not need to work in government to feel this. In the spring of 2020, New Jersey was still relying on unemployment systems more than forty years old, and its governor publicly appealed for volunteers who knew COBOL as claims overwhelmed them. You feel it when a permit for your kitchen takes longer than the kitchen. When a bridge repair is announced for a decade. When your doctor’s office still runs on fax, not because any law requires it, but because the systems on either end were never made to talk. At work, you feel it as the meeting held to prepare for the meeting, the report nobody reads, the decision postponed because the right five people were not all in the room. One company that tracked its own delays found it was postponing decisions six times in ten. A survey of workers in Australia, one of our closest allies, found ordinary staff spending more than six hours a week complying with rules their own organizations had imposed on themselves. That is the same plaque, in your week, and in your allies’ weeks.

Here is where the old arguments fail us. For forty years the debate has had two sides. One side says the machine is starved and needs more money and more people. The other says the machine is bloated and needs less of both. Both treat the machine as a budget. It is not a budget. It is a clock. Starve a slow machine and you get a slow, starved machine. Feed it and you get a slow, fed machine. Give it faster tools and you get a slow machine with longer queues. Neither side has asked a central question that changes the speed: how many hands must a decision pass through, and how many of those hands are still there for a reason anyone can name?

The honest answer is that we do not know, because we do not count it. Agencies report hundreds of performance measures by law. What the government lacks is a comparable ledger of decision cycle time, the days from a need being raised to a decision being made, kept the same way across agencies and published. We count dollars to the penny. We publish the budget of every agency, and there is no comparable government-wide ledger of their decision paths. And the plaque forms in exactly that darkness. Something goes wrong. Someone is blamed. A rule is added so it cannot go wrong that way again. The rule needs a check, the check needs a form, the form needs a reviewer, and the reviewer needs a sign-off from someone senior enough to be blamed next time. Every one of those additions was somebody’s reasonable answer to a real failure. None came with a date to leave. Every process we build gets a birth certificate. None of them gets a death certificate. That is a recurring mechanism, and it does not care which party is in power, because both parties have spent decades adding layers, each for a defensible reason, and neither has built a way to take them out.

Tending the system is less dramatic than cutting it, funding it, or buying it another tool, which is why nobody runs for office on it. It looks like this. Measure cycle time the way we measure spending, and publish it, by agency, by year, on a page anyone can read. Put the map of who must sign on the table, where the people who wait can see it, and make someone answer for its length. Give the heaviest internal procedures, the approval layers, and any rule without a named owner or a purpose anyone can state a mandatory review date, not to kill them on a schedule but to force the decision to keep them. Treat the operating model of a department the way we treat a bridge: an asset that wears, on a maintenance schedule, inspected before it fails rather than after. And point the new tools at the right target. The best use of a machine that can read a million documents is not to write a million more. It is to find the rules nobody can name a reason for, count the days each one costs, and put that count in front of the person who can remove it. Used that way, the fastest tools in history become the first ones that make the system faster instead of busier.

The goal is not a simpler government. Complex problems need complex institutions, and a nation of three hundred and forty million people will never be governed by a pamphlet. The goal is a navigable one, where the speed of the answer is a number someone owns, where a rule that has stopped earning its place can be found and removed, and where the machinery can move at the speed of need when need arrives, as it always does, without warning.

Many allied democracies face analogous pressures, although their institutions and their performance differ. When the OECD tells its members it is time for a regulatory reset, and is careful to say it means smarter rules rather than fewer protections, it shows that the problem is neither uniquely American nor inherently ideological, and it is talking to all of us. The countries that would like to see the free world lose have noticed that we are slow, and they are buying the same tools.

The threat is not that our institutions will fail. The threat is that they will keep working exactly as designed, at the speed they were designed for, with better and better tools, in a world that has stopped waiting. The plaque did not come from bad people. It will not leave on its own. But it can be cleared, and the systems underneath it are still worth saving, because they are the only systems we have. They just need someone to start counting the days.