What this covers

Jennifer Pahlka and Tyler Cowen discuss why American government fails to deliver on its intentions, tracing the problem not to insufficient resources or accountability mechanisms but to a fundamental mismatch between how government operates and how systems actually improve. Pahlka, drawing on experience at the U.S. Digital Service, argues that government runs on a waterfall model—a stepwise descent of law through bureaucracy—while effective systems anywhere else operate through tight feedback loops, tolerance for small-scale failure, and constant revision toward original intent. The core issue is that accountability to procedure has replaced accountability to outcomes; agencies meet targets without producing the results citizens need. Cowen adds a structural observation: the U.S. may possess unusual state capacity precisely because military strength requires the ability to command compliance, making America strong in ways that produce collateral damage elsewhere.

The conversation moves through several concrete territories. On procurement, Pahlka contends that government buys software as it buys pencils—five-year requirements cycles, bids, delivery—when software demands internal technical competence and continuous iteration; without insourcing this skill, the AI era will accelerate privatization of governance itself. On policy accretion, she notes unemployment insurance remains unworkable after ninety years of accumulated regulations layered atop the 1935 Social Security Act. On institutional resistance, she observes that even elite talent and leadership support cannot make large bureaucracies fundamentally more agile because the system itself fights back; the bureaucratic rigidity cycle creates new innovative organizations that lack power against entrenched procurement frameworks. She introduces stop energy versus go energy—the observation that reformers encounter roughly ten times as much blocking infrastructure as building infrastructure. Both speakers weigh whether disruption is necessary and discuss how Goodhart's Law has corroded government measurement, what distinguishes her vision from calls to simply shrink government, and which states are quietly outpacing others on permitting and procedural reform.

Sharpest takeaway

Pahlka argues that American government fails not from lack of resources or accountability but from a 'waterfall' industrial-era operating model with too much process-fidelity 'stop energy' relative to outcome-focused 'go energy'; fixing it requires iterative feedback loops, tolerance for small-scale failure, and in-house technical competence rather than orderly reform.

  • Accountability to procedure has displaced accountability to real outcomes (Goodhart's Law) in government
  • Excess resources and accumulated policy legacy, not scarcity, often produce worse service delivery
  • Disruption may be necessary because the bureaucratic system itself fights individual reformers

The argument · threads6 threads · 27 claims
0.70

Government's waterfall paradigm prevents feedback loops that allow laws to achieve their intended real-world effects.

4 pointscentrality 5/5
  • Government operates in a 'waterfall' paradigm — a one-way stepwise descent of a law through the bureaucratic hierarchy to implementation — whereas things that work well in real life are iterative, returning to original intent through tight feedback loops; moving from waterfall to an agile, feedback-closing model is a profound paradigm shift that will require substantial disruption.

    “The Waterfall process is, you do this thing and then it goes into this stepwise procedure and gets handed off to others. It’s really a one-way trip.”

  • Members of Congress increasingly, if secretly, recognize that the laws they pass do not produce the real-world effects they intend — they turn the steering wheel but it isn't connected to the wheels — and this sense of powerlessness, driven by the law of unintended consequences, has contributed to Congress's passivity.

    “they are turning the steering wheel, but it’s not connected to the wheels of the car. They’re pressing the gas, but the car’s not going anywhere.”

  • Punting governance decisions into the judiciary is harmful because tight feedback loops require acting quickly while courts act slowly; relying on litigation to achieve policy goals means harm is already done before lawsuits resolve, and the constant fear of being sued prevents iterative operation in the bureaucracy.

    “those tight feedback loops require you to act quickly, and the courts do not act quickly.”

  • Government should be judged not by the degree to which it can fail but by how quickly it can recover from failure; the lean startup approach of failing fast on a very small scale — working out bugs before rolling out to millions — produces more robust, resilient, and scalable systems than government's habit of trying to serve everybody perfectly on day one without prior testing.

    “we should not measure government by the degree to which it can fail, but more by how quickly it can recover”

0.70

Measuring fidelity to procedure rather than actual outcomes creates false accountability that erodes over time.

4 pointscentrality 5/5
  • Goodhart's Law — that once a measure becomes a target it ceases to be a useful measure — appears everywhere in government; the 1990s reinventing-government and new public management reliance on setting and following goals had real value but eroded as people met targets without meeting the intended outcomes, which is why a faster-iterating digital approach that can revise targets is needed.

    “Goodhart’s Law that says once a measure becomes a target, it ceases to be useful, and you see that everywhere in government.”

  • The way government structures accountability is flawed because public servants are held accountable to metrics that are not proxies for the real outcomes people care about; high focus on accountability that is really about fidelity to procedure and process rather than to actual outcomes is not genuine accountability at all.

    “we are holding public servants essentially accountable to metrics that are not proxies for real outcomes that people care about.”

  • To make a government like New York City achieve its goals, four things must change: fix the civil service hiring/firing system (NYC is stuck in Moses-era civil-service exams given only at certain times), reduce procedural bloat (which itself creates the ineffective hiring system), invest in digital infrastructure, and close the feedback loop between policymakers and implementers by stopping the 'fire and forget' approach and building affordances for ongoing iterative feedback.

    “you need to be able to hire the right people and fire the wrong ones... Second, reduce procedural bloat... third thing is you have to invest in digital infrastructure... the fourth thing is that you have to close that feedback loop between policy and implementation.”

  • America's social safety net is fundamentally paperwork — many low-value jobs across multiple levels of government consist of processing forms, IT systems, and compliance, when what a safety net should do is help a human being through a tough time; AGI should move workers from this low-value paperwork toward higher-value work that actually helps people, even though it will probably displace some workers.

    “All of our social safety net is paperwork. It’s lots and lots of different kinds of paperwork at lots of different levels.”

0.57

Smaller, faster, more constrained systems often deliver better outcomes than over-resourced bureaucracies built on procedural accretion.

8 pointscentrality 3/5
  • Working in the video game industry taught Pahlka that one can imagine fundamentally different worlds, economies, and systems, and that the constraints in our daily lives are manufactured rather than fixed — captured by David Graeber's line that the ultimate hidden truth of the world is that it is something we make and could just as easily make differently — which applies to the supposedly immutable civil service, since the US already has 50 wildly different state civil services to learn from.

    “You can have a whole different world, and once you realize that, all these constraints in our daily lives — you realize they’re manufactured.”

  • DOGE represents the disruption that was inevitably coming even if not the one that was wanted; Pahlka regrets that a more orderly attempt at responsible disruptive change was never tried, because past reform efforts (like reinventing government and Unit X) were too respectful of old institutions and 20th-century mindsets.

    “I do think this is the disruption that we’re getting, whether it’s the disruption we wanted... I just feel sad that we didn’t try.”

  • Government can be both smaller and better, but it is hard to tell whether current disruption is aimed at making government work better or merely making it smaller; Pahlka's own advocacy for a shock to the system comes from a desire for government to work, which she distinguishes from Vivek Ramaswamy's proposal to fire bureaucrats based on the last digit of their Social Security numbers.

    “Sometimes, it’s hard to see whether what’s going on right now is about government working better or just being smaller. I think it could be smaller and be better.”

  • Interesting governance change worth emulating is happening at the state level — Pennsylvania under Shapiro has worked to speed up permits (per a recent Don Moynihan paper) and Colorado under Polis is doing interesting work — because some states have either tackled procedural accretion or never let as much procedure accrue, unlike California.

    “I am very interested in Pennsylvania under Shapiro. Great paper by Don Moynihan recently about the work that they’ve done to just speed up things, like permits.”

  • American government has uniquely deep policy complexity and legacy because over roughly 250 years it has added and very rarely subtracted; unemployment insurance, which should be a simple program, still doesn't work and would be a disaster in another unemployment spike because the 1935 Social Security Act has accumulated some 90 years and thousands of pages of active regulations, so the challenge isn't digitization but building resilient, scalable systems atop that much accreted policy craft.

    “We didn’t clean up for 250 years. We’ve added and very rarely subtracted.”

  • When Pahlka was finishing her book, the Niskanen Center was the only think tank doing work on state capacity; she now works there building a team to articulate how any US government at any level can increase its ability to achieve policy goals, drawing interest from people who never used to attend to technical infrastructure in agencies, including health staffers and members of Congress.

    “They were the only think tank that had anything on state capacity when I was finishing my book.”

  • New York City is cleaner and safer than it was in the 1980s, but probably because it is richer rather than because it is run better.

    “It’s certainly cleaner and safer than it was in the ’80s when I was growing up.”

  • Counterintuitively, situations with less money and fewer people often deliver services better; California, despite being tech-forward and well-funded, had the second-lowest SNAP participation rate in the country because abundant money led it to over-spec the enrollment system, gather too many requirements, hire expensive vendors, and produce a 212-question form that timed out before applicants could finish — whereas Maine spent orders of magnitude less and enrolled far more.

    “Because we could put a ton of money into it, we way overspec’d it, gathered way too many requirements, gave it to expensive vendors, and we ended up with an enrollment system that had 212 questions.”

0.56

Entrenched institutional systems resist change even when individuals champion better practices and achieve local success.

4 pointscentrality 4/5
  • Even with elite talent, strong leadership support, and good local outcomes, large institutions like the Department of Defense do not become fundamentally more agile, because the system itself fights back against any individual trying to change it — the problem is not merely that people resist change, since most people want change.

    “somehow the system itself fights back against any individual who’s trying to change it.”

  • In any agency there is a small amount of 'go energy' — people building something better, cheaper, and faster like the Direct File team — surrounded by roughly ten times as much 'stop energy' (infrastructure, people, rules that block them); some stop energy is appropriate, but the balance is badly off, so reformers seeking effectiveness must identify and reduce the stop energy rather than cut the go energy first.

    “Around them, you have 10 times as much infrastructure, people, rules that are stopping them. That’s the stop energy.”

  • Government repeatedly creates new innovative organizations to bypass old ones (the 'bureaucratic rigidity cycle'), but has not been disciplined about dismantling the old when creating the new, so new skunkworks like USDS end up with far less power than existing CIOs and procurement frameworks; if you only add new layers without disrupting the old, the disruption fails.

    “That’s what we call the bureaucratic rigidity cycle. We haven’t, I think, been disciplined enough about dismantling the old when we create the new.”

  • An Air Force official argued defense budget cuts need to be on the order of 50 percent because the historical 15-20 percent cuts don't shock the system enough to actually change behavior; only a large enough shock — like a hot war — forces real behavioral change in entrenched institutions.

    “it needs to be 50 percent because when we’ve had 15 percent or 20 percent cuts, it doesn’t shock the system enough to actually change behavior.”

0.49

Government must insource software and AI competence rather than procure it like commodities to avoid privatization.

2 pointscentrality 4/5
  • Government's core procurement problem is that it buys software the way it buys pencils and cars — spending five years developing requirements, sending it out for bid, receiving it, and declaring the thing bought — when software and AI require internal competence to run a far better and faster process and to own and operate the software continuously; without insourcing this competence, the AI era will drive further privatization of both software and operations.

    “The problem with our procurement today is that we buy software the way we buy pencils and cars. We’ve bought a thing. AI isn’t like that, and actually, software isn’t like that.”

  • AI is the first transformational technology that did not fundamentally come out of US government resources — it cannot be traced back to DARPA or DOD investments the way prior commercialized technologies could — which means government does not really understand it and faces an unprecedented question about the relationship between AI companies and the US government itself.

    “this is the first time that such a transformational technology hasn’t fundamentally come out of US government resources. This is not one that you can ultimately trace back to DARPA or DOD.”

0.49

Civil service protections must preserve expertise and scientific integrity against wholesale purges between administrations.

2 pointscentrality 4/5
  • An independent civil service must be protected from firing for certain reasons so it can tell an administration what the science actually says even if the administration chooses not to follow it, retaining scientific and bureaucratic expertise; without this, the entire civil service risks being wiped out with every change of administration, which is the trajectory we appear headed toward.

    “We do want an independent civil service that can say, “Actually, the science says X.””

  • Public-sector unions often advocate policies not in their members' interests — notably protecting the small number of genuine underperformers who drive the rest of the public service crazy — when for the good of most public servants they should make it easier to remove those underperforming employees; this persists because the costs of protectionism are widely and indirectly distributed while a fired worker screams loudly (asymmetric information).

    “It’s crazy to me that the unions don’t say, “Actually, for the good of most public servants, let’s make it easier to remove these underperforming employees.””